Why Census Shows Fewer Men

The UK has 800,000 fewer young men than previously thought. This pattern was originally identified in the 1991 census, but given a lack of confidence in the follow up survey for that census, the numbers were revised to restore the predicted pattern. In 2001, the pattern has been confirmed and validated by the one number census. We will revise population estimates back to 1982. The critical factor appears to be emigration. The International Passenger Survey works well, but it captures travellers' intentions at the time of departure. These may be prone to change once people are abroad, particularly among young men with few ties at home.

In England and Wales, the census shows that the actual population is some 900,000 lower than previously estimated by the mid-year estimates.

The difference is made up almost entirely of young men. The ratio of men to women is lower, at a younger age, than had been estimated previously. The census shows that women are now already outnumbering men in the 20-24 age group.

The pattern of sex ratios found in the 2001 census confirms that found in the 1991 census, this is shown in Table 2. At the time of the 1991 census these findings were discounted given the problems with the post-enumeration survey. It was considered at that time that there was no credible evidence to support the observed pattern. Alternative sex ratios, based on 1971 and 1981 census results, were adopted in the mid-1991 population estimates (for more details, see 1991 CVS coverage report ISBN 0 11 691591 9). The 2001 census results together with the census Coverage Survey (CCS) provide a consistency which lends credible support to the sex structure identified in 2001 and implies that revisions also now need to be made to the population benchmark for 1991 and hence all years from 1982 to 2000.

The pattern is spread across the country
The pattern of men to women in the 2001 census estimates is seen widely across the country and not just in particular types of places. For example, almost 75 per cent of the 376 local government districts in England and Wales have more females than males in the key age group 20-39. These districts range in size and socio-demographic composition and are not clustered in particular parts of the country. Many of the other 25 per cent of local government districts in England and Wales are places where there are particular reasons for there to be more men than women in the key age group 20-39. For instance, they include areas with a large armed forces population.

Changes at local authority level
Revisions to population estimates will be made back to 1982 to take on board the new benchmark results from the 2001 census. To give a guide to the revisions that will be required at local authority level the linked table shows for England and Wales a comparison between the 2001 census results and the current mid-year population estimates for 2000.

Rebasing the population estimates
The mid-year estimates are the best estimates that can be made annually on a nationally consistent basis but inevitably they are subject to error which can accumulate between benchmarks. To give an idea of the scale of the revision that will now be needed to the estimates the reduction of 1.7 per cent at national level is equivalent to an average error in each year of 0.1 per cent. This section includes a discussion of the sources of error in the MYEs and some of the issues surrounding the extent of the revisions that will be required.

Migration in the MYE

In the UK, the mid-year population estimates are based on an accepted and widely employed demographic method, the cohort component method. Using this method, estimates are produced by updating from a census base allowing for the births, deaths and net migration that has occurred since the census. A detailed description of the method used to produce population estimates during the intercensal period is available on the National Statistics website.

The most difficult component of population change to estimate accurately is migration. There is no comprehensive registration of migration in the UK, either of moves to or from the rest of the world, or of moves within the UK. Thus estimates of migration have to be based on survey data and the best proxy data that exist. This is in contrast to the estimation of natural change which is the other major component of population change. Natural change is the excess of births over deaths and this is estimated using civil registration data. There is compulsory registration of births and deaths in England and Wales and there are clear advantages to registering these events. It is generally accepted that these data are likely to be virtually complete.

Any errors in estimating annual population change due to migration will remain in the estimates and will be added to the errors in subsequent mid-years. Thus, the mid-2000 population estimates will have a wider error band than earlier mid-year estimates during the 90s, which incorporate fewer years of estimated population change. Errors in estimating migration are not the only reason for the differences that exist between the mid-2000 and the 2001 census population estimates. Another source of error in the mid-2000 estimates is from the adjustments that were made to the census base for under-enumeration in the 1991 census.

What happened in the 1991 census
In usual circumstances, it would be expected that the estimates for mid-1991 to mid-2000 would be based on the 1991 census together with allowance for under-enumeration in that census. However, there were difficulties with the assessment of under-enumeration in the 1991 census. The 1991 post enumeration survey (the Census Validation Survey, CVS) was a combined quality and coverage survey in which a sample of about 6,000 households were selected for interview. It was not completely independent of the census and it was carried out to the same basic procedure as the original census. Therefore, any limitations inherent in the census would remain in the CVS. In particular, there may have been some people who were not counted by either the census or the CVS.

The 2001 Census Coverage Survey (CCS) is a big improvement on the 1991 CVS in many ways. The 2001 CCS was a large survey of just over 320,000 households, it was conducted independently of the census and it was successful in identifying individuals who were missed in the census. This article provides more information about the 2001 CCS: Pereira R (2002). The Census Coverage Survey - the key element of a One Number Census. Population Trends 108

Given the problems with the 1991 CVS, it was decided that the total figure for the national population of England and Wales in mid-91 would be a number that was very close to that derived by updating from the 1981 census. This was judged to be more plausible than the national total based on the results of the 1991 census. It was only this one figure that was used - the national total of the rolled-forward from 1981 estimates. The 1991 census results were used to help derive the area distribution and the age and sex profile. It was necessary to adjust the mid-91 estimates by age and sex, otherwise they would not sum to the agreed national total.

Total was adjusted for lack of young men
Thus the census and CVS results had to be adjusted for under-enumeration that had not been identified by the CVS. To achieve this England and Wales was divided into 8 area types and differential adjustments for under-enumeration were made by area type, sex and age group. Certain of the area types, namely inner London, outer London, metropolitan cities, and some other non-metropolitan cities, were given quite large adjustments, particularly to add in more men aged 20-29. In addition to the area type adjustments that were made in 1991, some other population subgroups were also enhanced using administrative data sources to help obtain a full count. These groups were babies, armed forces personnel and older people.

The availability of results from the 2001 Census provides the first opportunity to review the adjustments for under-enumeration in the mid-1991 population estimates. It is now clear that information on age sex ratios observed in 1991 have been largely confirmed by the 2001 census and census coverage survey. This suggests that the mid-1991 adjustments for under-enumeration and, by implication, the national total updated from 1981, were not, in fact, correct and that there were accumulated errors in the population estimate based on the 1981 census results.

Estimates were too high
Based on the evidence now available, it is estimated that the mid-1991 estimates were too high by about 300,000 people. In view of these findings, the population estimates for mid-1991 will be reconstructed, taking this discrepancy into account. The methodology for doing this will be subject to external expert review and will be described in a paper on the National Statistics website before the end of the year.

Reconstructing the mid-1991 estimates means that it will also be necessary to revise the population estimates for the 1980's as well as the 1990's. It is likely that this work will be carried out as follows. First, the 1991 base will be reconstructed. Next, the population estimates for mid-1992 to mid-2000 will be revised so that they are consistent with both the 2001 ONC and the reconstructed 1991 base. Finally, the population estimates for mid-1982 to mid-1990 will be revised so that they are consistent with the reconstructed 1991 base and the 1981 base. More details about the methods for revising the intercensal series can be found in a recent Population Trends paper. The revised population estimates for mid-1991 to mid-2000 will be published on 13th February 2003 and the earlier revised series for mid-1982 to mid-1990 will be available in mid-March 2003.

Population Change between 1991 and 2001
Table 3 below shows the estimation of population change between 1991 and 2000 based on previous censuses and, for comparison, the population change between 1991 and 2001 based on the results of the 2001 census. This table shows the impact on the migration estimates of the 2001 census estimate, including the adjustment to the mid-1991 base. The estimate of net migration is substantially less than has been estimated during the 90s. Natural change is slightly higher in the second part of this table purely because it covers a slightly longer time period. Part a of the table covers change over the 9 years between mid-91 and mid-2000, part b covers a longer period of the 9 years and 43 weeks between mid-91 and census day 2001.

Errors in estimating causes of change
As has already been explained estimating migration is the most difficult component of population change to get right. There are two broad types of migration, international which is between England and Wales and the rest of the world, and internal which is within the UK, including cross-border flows between England and Wales and the other constituent countries of the UK. So far, investigations have been focussed mainly on international migration but further work is planned to help assess the accuracy of the internal migration flows and the results of that work will be reported later, early in 2003.

There are two major strands to the international migration estimates. These are the estimates of in and out migration that are survey based and which are derived from the International Passenger Survey (IPS). Added to these flows are data from the Home Office in respect of immigrants who have not been identified by the IPS. These include most asylum seekers and other people who apply to the Home Office for permission to remain in the country having initially entered for a short stay (these people are referred to as visitor switchers). Thus, the estimates of immigration are compiled from a number of sources. We have no evidence at this stage to indicate that immigration has contributed to the revision now needed to the current population estimates although it is recognised that statistics on immigrants are difficult to compile and quality assure with confidence.

There is less information about outward migration and, in the past, we have used the IPS on its own to produce complete estimates of emigration. The results from the 2001 census suggest that this approach may have led to an underestimation of out migration. More work is needed to identify other data sources that

Table 4 shows the components of population change. This illustrates that net migration is the difference between estimated gross in and out migration flows which are large. Small changes in the gross flows can have a big impact on the estimate of net flows.

Estimating migration
The main data source used to estimate flows between England and Wales and the rest of the world is the International Passenger Survey (IPS). This link provides more information about the IPS and estimates of International Migration see (link to International Migration Volume). The IPS is the richest source of information on international migration into and out of the UK. It is a continuous voluntary sample survey, based on face-to-face interviews, which covers the principal air, sea and channel tunnel routes between the UK and countries outside the British Isles.

The IPS is a well designed survey that provides valuable data. But a potential source of over or underestimation in the IPS arises from the fact that it measures intentions which may not be realised. A person who comes here stating an intention to migrate who then subsequently leaves after only a short stay would not be counted out as a migrant. On the other hand, the opposite case where people state their intention is to visit and then stay longer thus becoming a migrant to England and Wales, is at least partly allowed for by the addition of Home Office data in respect of visitor switchers. Another difficulty arises where people leaving state their intention is to be away for a short time but then they do not return.

Evidence of UK emigration
There is evidence to suggest that the international migration data sources have not been capturing all out migration. Data from a selection of key recipient countries which include Australia, Canada, New Zealand and the USA show that there are significant numbers of UK born people among immigrants to those countries. Most of these emigrants from the UK are young to middle aged adult men. Due to definitional differences and restrictions on the data that are available, it is not easy to carry out comparisons of the data from these countries with the estimates that have been used in England and Wales.

One country where there is some definitional similarity, making comparisons possible, is Australia. Data from the Australian immigration control system shows that for the period for 1981 onwards there has been a net flow of approximately 260,000 permanent and long-term settlers to Australia of people born in the UK. Our combined data sources estimate 46,000 less than this. In addition, there has been a large increase in short-term arrivals in Australia by people born in the UK aged 25-44; these people are also more likely to be men than women. Some of these short-term visitors may stay on. Thus there is a strong likelihood that the underestimate of migration to Australia exceeds the 46,000 that can be measured from the permanent and long-term flows.

In short, then, evidence does exist from recipient countries to support the proposition that there is out migration, particularly of young to middle-aged adult males that is not being captured by the data sources that have in the past been used to estimate migration. Similar analyses are not readily possible for other countries. It may well be that what is seen for Australia will also be seen for other countries.

Between 1991 and 2000, some 200,000 people have been added into the mid-year population estimates having been given extended leave to remain in the UK. These were people who came from countries other than the USA, Europe or the Old Commonwealth. Visitor switching from these other countries was assumed to net to zero, i.e. people switching from visitor to resident status here would be balanced by people switching out from the UK to become residents elsewhere. During the 80s, it was assumed that all visitor switching balanced and no additions were made to the population estimates in respect of visitor switchers. If this approach had been taken during the 90s there would have been 200,000 less in mid-2000.

New approach to 'switchers'
It may be that we need to adopt a more conservative estimate in the future and we intend to develop an appropriate method in consultation with HO. In the meantime for the 9 weeks between census day and the mid-year an assumption has been made that the total number switching to residence status in the UK is balanced by outward visitor switching of UK residents. This is in line with the assumption used during the 80s. Thus no visitor switchers have been included in the mid-2001 population estimates. If an allowance had been made for visitor switchers, the numbers would be relatively small and would have little impact when spread between local areas. If it is later determined that some allowance should be made for inward visitor switching these will be incorporated in the base used for future estimates, mid-2001 will not be revised.

The IPS is subject to both sampling and non-sampling error which are difficult to quantify precisely and investigations are on-going. These, and other, issues are currently under consideration as part of a National Statistics Quality Review (NSQR) on International Migration. This review provides an ideal opportunity and framework for making recommendations for improving international migration data sources over the coming decade. In addition, as previously explained, ONS will be conducting an assessment of accuracy of the mid-year population estimates. An important aspect of the assessment of accuracy of the MYEs and the NSQR, will be to assess the sufficiency of the information that is used currently to supplement the IPS and to plan ways in which it might be improved in the future. As explained at the end of this commentary, updates on progress will be given at the Liaison Group on Population Statistics.

The concept of usual residence
The 2001 census results are based on the One Number Census methodology incorporating full allowance for all those missed by the census. They provide the best possible population base for subsequent estimation of the resident population. A traditional census would have produced far less robust estimates in 2001 of those usually resident in an area. The issues we have had to grapple with in conducting the census have shown that there are increasingly complex issues in defining the concepts required for understanding the population and for measuring those concepts robustly.

There have been big changes in society in the last 10-20 years and one way in which this is manifesting itself is in where people consider themselves to live. For instance, some people have become more mobile since 1991, travelling around the world and staying temporarily in different places. Results becoming available from the 2001 census provide an exciting opportunity to learn something about these societal changes and their implications for population statistics. Our future development programme will take this opportunity in order to ensure that our system of population statistics in the future remains robust and relevant.

Reassessing population estimates
ONS will use this unique opportunity to assess the accuracy of the mid-year estimates that were made during the 90s, to learn lessons and where we can to improve our methods. Some ideas for further development are outlined here. Updates on progress to implementing these improvements will be given at the Liaison Group on Population Statistics, which is the advisory group on population matters which exists under the aegis of the Central and Local Government information partnership (CLIP).

ONS has already established a research programme to explore how census-type information, and data for small area social statistics in general, could be collected in future. The scope is very wide. As well as exploring modifications to a traditional census, taking account of the lessons from the 2001 census and harnessing modern technologies such as the internet, other potential options will be explored. These include: a rolling census, covering successive sections of the population; a sample census, only ever covering a percentage of the population; producing statistics from existing person-based data held by government for administrative purposes; some combination of these.

The new insights about residence and international mobility will be relevant to all these future options. The first phase of the research will result in recommendations in late 2003, including a development and testing strategy to be implemented over the subsequent years. ONS will liaise with other government departments and will be consulting with key stakeholders including Census Advisory Groups during the research process.