Census methods used in other countries

Survey of census methods for the 2010 round of censuses

Each decade the United Nations Statistics Division (UNSD) conducts a world-wide survey of country plans for the next round of censuses. For the 2010 census round, the UNSD conducted the survey in June 2009 and published the results in February 2011 (United Nations 2011b).

Of the 233 countries that were sent a questionnaire, 138 countries (59%) responded. There was considerable regional variation in response rates, with response rates exceeding 75% in Asia, Europe and Northern America, but only reaching 20% in Oceania. Table 1 presents a summary of the results for the questions on the census methods used.^{Footnote 1}

Of the 138 countries that responded to the survey, 115 (83%) use the traditional census as the main source of data to produce their population count, 15 countries (11%) use administrative registers as the main source, 1 country (France) uses a rolling census, and 7 countries (5%) use other methods. These 'other' countries are primarily countries where administrative registers are used for some census activities, but are not used as the main source of census data. Of the 15 countries using administrative registers as the main source of census data, 12 are in Europe, 2 (Bahrain and Singapore) are in Asia and 1 (Greenland) is in Northern America.

In addition to the main census method, 49 of the 138 responding countries report using additional sources to provide data on specific census topics. Of the 115 countries that conduct a traditional census as the main method, 31 (27%) report using other sources of data, including 20 that use annual, regular or ad hoc sample surveys and 17 that use administrative registers. The latter could include, for example, obtaining data for some data topics from administrative sources, as Canada did for income data using tax files in the 2006 Census. Among the 15 countries that use administrative registers as the main source, 10 (67%) report using other sources, including 6 that use an annual or regular sample survey and 4 that use an ad hoc sample survey conducted specifically for the census. The other 5 countries, all of them in Europe, use administrative registers exclusively for their census.

For the UNECE region more specifically, 50 of the 56 countries responded to the survey, and more precise details are available on the methods that are used. Table 2 lists the UNECE countries that are using a traditional census, a register-based census combined with complete enumeration and/or surveys, a register-based only census, and other census methods for the 2010 round.^{Footnote 2}

Two other non-UNECE countries to which Canada is often compared are Australia and New Zealand; both are using the traditional census method for their 2011 censuses.^{Footnote 3}

While the traditional census remains the predominant method for the 2010 round, there is a definite shift in the UNECE region towards using registers in census-taking. Table 3 shows the number of countries by census type in the 2000 and 2010 rounds.

Of the 37 countries that conducted a traditional census in 2000, 7 countries (Czech Republic, Estonia, Israel, Italy, Lithuania, Poland and Turkey) now plan to use administrative registers in combination with complete enumeration and/or surveys. One country (Austria) is planning to move to a register-based only census, and 2 countries (France and the United States) are now using other methods.

Of the 6 countries that used a combined approach and the 3 countries that used a register-based only census in 2000, all of them plan to use the same approach in the 2010 round.

Finally, of the 4 countries that did not conduct a census in the 2000 round, Bosnia and Herzegovina is planning to conduct a traditional census; Germany and Iceland are planning to conduct combined censuses; and Sweden is planning to conduct a register-based only census.

Experiences of selected countries

Survey of census methods for the 2010 round of censuses provided a statistical overview of the census-taking methods being used both worldwide and in the UNECE region in the 2010 census round, as well as general trends in the UNECE region since 2000. In the course of the literature review, documents describing the experiences of specific countries in planning for their 2010 Census or beyond were also examined. This section summarizes the experiences of selected countries that are contemplating major changes for the future (United Kingdom, New Zealand, Australia and India) or that have made major changes for their recent censuses, either by employing administrative registers (Nordic countries, Netherlands, Germany, Switzerland and Austria) or by adopting a continuous survey collection approach (United States and France).

The descriptions that follow are not intended to cover all of the possible variations in census-taking methodologies or to imply that there is nothing to be learned from countries not described. Rather, the goal is to illustrate the major features of census-taking approaches as applied in practice, as context for the assessment to follow in Assessment of census approaches in the Canadian context.

United Kingdom

The United Kingdom has carried out a traditional census every 10 years since 1801. Census Day in 2011 was March 27.

In October 2003, the Office for National Statistics (ONS) published a discussion paper entitled Proposals for an Integrated Population Statistics System (ONS 2003a).The discussion paper was accompanied by several supporting papers on related topics, including linkage of existing data sources (ONS 2003b), the use of the rolling census approach (ONS 2003c) and a review of international approaches (ONS 2003d). The main drivers behind the proposals were increasing non-response rates with the traditional approach and the increasing costs associated with this trend, as well as the need for more frequent and timely data for small areas and subgroups.

The proposals identified five key areas of work that would have to be brought together (ONS 2003a, page 6):

• the creation of a high-quality address register
• the registrar general's feasibility study into the development of a population register
• administrative data linkage
• the 2011 Census
• forming a continuous population survey from the existing range of public household surveys.

The hope was that the new Integrated Population Statistics System (IPSS) would be in place by 2013, replacing the traditional census. Comments were solicited on the proposal, and the results were published by ONS in 2005 (ONS 2005a). The ONS summarized the comments received as follows:

In general, users generally expressed strong support for the overall proposals for an integrated population statistics system (IPSS), which were described as exciting and visionary. However, nearly all the comments emphasised what a massive undertaking it would be, and a common concern amongst users was that the strategy was high risk and that the proposals did not perhaps fully recognise the risks. Nevertheless, the feeling was that the vision was worth striving for.

The feasibility study into the creation of a national population register was an ONS-led effort known as the Citizen Information Project (CIP). In April 2006, the CIP was folded into the National Identity Register, which was the database underlying the Identity Cards Act 2006. The Act and its associated U.K. National Identity Card subsequently became controversial, especially following the October 2007 loss of some 25 million records of child benefit recipients. In January 2011, the Conservative-Liberal Democrat Coalition government repealed the Identity Cards Act 2006, and all records in the National Identity Register were to be destroyed by February 2011.^{Footnote 4}

In April 2009 the ONS commissioned survey research into the public's views on data sharing between government departments and the creation of a single central database to hold basic personal information (ONS 2009). In response to a question on the creation of a single central database, 64% of respondents were supportive, 13% neither supported nor opposed the idea, 18% were opposed and 5% didn't know.

The ONS is now looking at options beyond its 2011 Census (UNECE 2010b). The ONS has identified three broad categories of options: a point-in-time census providing a population 'spine,' administrative data as the basis for population estimates, or a survey or rolling census approach. These are very similar to the approaches considered in this paper. The ONS expects that it will complete its work on 2021 options development in 2014, and stresses that "…the work on alternative options does not imply a decision on a traditional 2021 Census."

New Zealand

New Zealand has a statistical system very similar to that of Canada, although its population is considerably smaller. Like Canada, New Zealand conducts a mandatory census every five years to provide population counts and socio-economic data, and to serve as the basis for other statistical programs. The 2011 New Zealand Census was to consist of a dwelling form containing 22 questions and an individual form containing 48 questions. Sampling is not used (i.e., there is no 'short' form).

Statistics New Zealand has examined the feasibility of moving to a population register-based approach for future censuses (Bycroft 2010). The main driver was the escalating costs of conducting a census; response burden is not considered a major issue in New Zealand.

The study concluded that New Zealand currently meets none of the conditions^{Footnote 5} for introducing a register-based census. The legislative framework, while it partially meets the requirements for using administrative registers, would require changes to privacy legislation to permit the long-term retention of unique identifiers and data linkage on a regular rather than case-by-case basis. Second, there has been no public discussion of the rationale and benefits of using registers. The paper describes the cultural aspects of this debate, citing the public arguments in the United Kingdom concerning the introduction of the U.K. National Identity Card. Third, New Zealand has no common unique identifier in use across administrative systems that could be used to link data sources, and in fact, such an identifier is expressly forbidden by the Privacy Act. Finally, there is currently no population register in New Zealand.

The paper concludes that the creation of a population register would have to be a long-term government initiative, and that two or three censuses would likely elapse before it could be used. While it did not carry out a formal comparison of the costs of conducting the census to the costs of establishing and maintaining a population register, it does state that there would also have to be strong administrative purposes to be served, and that countries with population registers "…see major benefits for government administration and efficiency and in having a rigorous system for personal identification." Public acceptability would be a key factor, as would the quality of the population register and other data that could be linked.

Australia

Like New Zealand and Canada, Australia conducts a mandatory census every five years. The 2011 Australian Census will be held on August 9. For 2011, as for 2006, the census questionnaire contains a total of 60 questions, covering both dwelling and person characteristics. All households are required to fill out the same questionnaire (i.e., there is no 'short' form).

Australia is currently beginning to plan its 2016 Census, which will again be a traditional census. Because of increasing demands for new census content, the Australian Bureau of Statistics (ABS) is looking at ways of increasing the number of questions included on the census, but without increasing respondent burden. One proposal for 2016 (Bell and Whiting 2009) is to have three different versions of the census questionnaire. All three versions would have the same 50 'core' questions, but each version would have a different set of 10 'thematic' questions. Each version of the questionnaire would go to one third of Australian households. An individual household would therefore only have to answer 60 questions, as in previous censuses, but data could be produced for a total of 80 questions—50 questions on a 100% basis and 30 questions on a 33% sample basis.

One disadvantage of the proposed method is that the thematic questions could only be cross-tabulated if they appear on the same version of the questionnaire. Because of this, the 30 questions would be divided into 3 themes of 10 questions each, with related questions that would naturally be cross-tabulated appearing in the same theme. As well, new methods of estimation would need to be developed.

At this juncture, the ABS has made no final decisions about the questions to be included or the methods to be used for the 2016 Australian Census. The ABS will begin conducting user consultations in 2012 to determine the 2016 Census content.

India

India conducted its first census in 1872, and since 1881, it has conducted a census every 10 years without interruption. The 2011 Census was the 15th in its history (Chandramouli 2009, 2010 and Government of India 2011). The legal basis for the census is The Census Act, 1948 and Census Rules, 1990, which specify that response is compulsory and that all data are to be kept confidential. The 2011 Census was conducted using the face-to-face interview methodology, using paper forms printed in 16 different languages. It involved 2.5 million enumerators and supervisors.

The 2011 Census was conducted in two phases. Between April and September 2010, data collection for the Houselisting and Housing Census was carried out. The houselisting and housing schedule contained 35 questions, including the total number of persons in the household, name and sex of the household head, condition of the building, number of rooms, and various household facilities (e.g., latrine, telephone, and drinking water). The second phase was the population enumeration, conducted during a three-week period from February 9 to 28, 2011. The population form contained 29 questions for each person, covering basic demographic characteristics, mother tongue and other languages known, education, economic activity (e.g., employed, occupation), place of birth, travel to work, migration, fertility and mortality, religion and scheduled caste or tribe.

Following data collection, the paper forms were scanned and the information was extracted using intelligent character recognition software. Provisional results for the 2011 Census of India were released on March 31, 2011, and showed that the population was approximately 1.2 billion persons. Final 2011 Census results are expected in 2012.

At the same time as the 2010 data collection for the Houselisting and Housing Census took place, the first phase of data collection for a new National Population Register (NPR) was conducted.^{Footnote 6} The NPR form was separate, and included 14 questions for each person: name of the person and resident status, name of the person as it should appear in the NPR, relationship to head, sex, date of birth, marital status, educational qualification, occupation/activity, names of father, mother and spouse, place of birth, nationality as declared, present address of usual residence, duration of stay at present address and permanent residential address. In a second phase of collection for the NPR, biometric data (photograph, fingerprints and possibly an iris scan) were collected for all usual residents of 15 years of age or more.

After several stages of validation and correction, the NPR data are forwarded to the Unique Identification Authority of India, who 'de-duplicates' the records using the biometric information and issues a unique 12-digit identification number for each person.^{Footnote 7} The cleaned information is then returned to the Office of the Registrar General and Census Commissioner, India (ORG & CCI), and will be used to form the National Population Register. The information collected will be used by the ORG & CCI to issue national identity cards to the entire population on a phased basis over the next several years.

The main purpose of the NPR is to create a comprehensive database of all usual residents (as defined by a six-month residency rule), in order to better target government benefits and services, to improve planning and to strengthen security. The legal basis for the NPR and national identity cards is The Citizenship Act, 1955, and The Citizenship (Registration of Citizens and Issue of National Identity Cards) Rules, 2003.

The 2011 Census is expected to cost some 22 billion rupees (approximately $481 million Canadian) and the creation of the NPR is expected to cost 37.6 billion rupees (approximately $820 million Canadian). The potential role of the NPR in future Indian censuses has been raised, but at this point no decision has been taken for 2021. Among the issues that would need to be addressed are the content of the NPR versus that of the census (the census generates much more detail) and the confidentiality of the information (by its nature the NPR is not confidential). One possibility under consideration is to combine the NPR with a regular house-to-house census.

Nordic countries

The primary experience with register-based censuses is in the Nordic countries. Table 4 summarizes the experience of the four Nordic countries that plan to implement a totally register-based system by 2011 (reproduced from UNECE 2007, page 5). In some cases, the table reflects plans as of 2007, as indicated by dates in italics or with question marks.

All four of the Nordic countries established Central Population Registers in the 1960s (1964 to 1969). Such registers evolved from local and national registers that in some cases date back several centuries. Three countries (Finland, Norway and Sweden) began using the Central Population Registers in their censuses of the 1970s, but it generally took several decades until these four countries were able to base their censuses totally on registers.

Denmark was the first country to adopt a register-only census approach. A Central Population Register, a Family Register and a Household Register were established in 1968, followed by the establishment of various other registers during the 1970s. All of these registers were used for the first time in a completely register-based census in 1981.

Finland was the next country to adopt a register-only census approach. It established a Central Population Register and an Income Register in 1969, and used these registers in its 1970 Census. Additional registers were established in the 1970s and 1980s, and the first register-only census was conducted in 1990.

Norway established its Central Population Register in 1964 and used it for the first time in its 1970 Census. It will conduct its first register-only census of population and housing in 2011, although it did conduct a register-only census of population in 2001, with additional field operations to collect information on housing and household membership, and to create a dwelling register.

Sweden established its Central Population Register in 1967 and used it for the first time in its 1975 Census. Although the Swedish Parliament passed legislation in 1995 mandating that the next census would be register-based only, it took until 2007 before the necessary legislation to construct a register of dwellings was passed (Axelson et al. 2010). Sweden will conduct a register-only census for the first time in 2011. It did not conduct a census in the 2000 round, although it did produce some of the data tables required by the European Union (EU) in 2001.

The experience of the Nordic countries suggests that once a Central Population Register is established, it can be used in the census within 5 to 10 years, but that the complete replacement of the census by registers takes much longer. In the case of these four countries, it has taken between 13 and 47 years from the time the Central Population Register is established to the complete changeover. The general pattern is that the Central Population Register is adopted first, and is initially supplemented by surveys or complete enumeration to collect information not available from the register. Then, as additional registers are developed and their quality becomes sufficient for census purposes, the collection of data is gradually replaced by administrative registers, until such time as field collection can be eliminated completely.

Netherlands

The first Dutch census was held in 1795, during a period when the country was a republic. After the Netherlands became a kingdom, censuses were conducted at approximately 10-year intervals, beginning in 1829. In 1859, the results of the Dutch census were used to set up municipal population registers, which were then maintained manually. For several censuses thereafter, the purpose of the census was twofold: to update and correct the municipal population registers, and to produce demographic and socioeconomic data on the population (Schulte Nordholt 2009).

The last traditional census in the Netherlands was conducted in 1971. Preparations for the 1981 Census showed a sharp decline in the willingness of the population to participate in the traditional census, resulting in a decision by the Dutch Parliament to postpone, and then eventually cancel, the 1981 Census. In 1991, the Census Act of the Netherlands was officially repealed (Bethlehem 2010).

However, due to its obligations to the EU, the Netherlands still had to conduct a census. For 1981, the census was replaced by tabulations from the central Population Register for demographic variables. The Population Register links the various municipal registers, which are kept up to date by the municipalities^{Footnote 8} and maintained using common software approved by the national government. For socioeconomic variables, such as labour and education, census-type data were produced from the Labour Force Survey (LFS), whose sampling fraction was increased to 5% for this purpose. The same procedure was used in 1991, although the LFS sampling fraction was only 1%. Housing data were produced using the Housing Demand Survey 1989/1990. Both the 1981 and 1991 censuses were extremely limited in output; for example, no data were produced at the municipal level. As well, there were inconsistencies in the data produced from different sources.

For 2001, the increased demand for data by the EU and other international organizations, as well as improvements in the quality of several administrative registers, led to the development of the so-called 'Dutch Virtual Census.' Because the content of administrative registers is not as rich as in the four Nordic countries presented in Census Methods Used in Nordic countries, the Dutch Virtual Census is based on combining administrative registers with existing sample surveys.

Two important methodological developments were essential to the virtual census approach (Linder 2004). The first was the development of a Social Statistics Database (SSD), created by linking and integrating data at the micro-level from the central Population Register, from other administrative registers covering topics such as employment, benefits and income, and from sample surveys such as the 2000 and 2001 Labour Force Surveys and the Survey on Employment and Earnings to provide data on variables such as education and occupation. The Population Register serves as the 'backbone' of the SSD. For the administrative registers, records could be linked to the Population Register using the unique Social Security and Fiscal Number (replaced in 2007 by a Citizen Services Number^{Footnote 9}). However, the sample surveys did not contain such an identifier; linkage was therefore done using date of birth, sex and address, occasionally resulting in some mismatches or non-matches. Once the data were linked at the micro level, the data were integrated by applying rules to resolve discrepancies.

The second major development was that of repeated weighting (Gouweleeuw and Hartgers 2004, Houbiers 2004). Variables that appeared on the registers were simply tabulated directly from the registers (i.e., they represented direct counts). However, for variables that appeared only on a sample survey, estimates were prepared using the technique of repeated weighting. This method ensures consistency of any new tables with all tables produced previously.^{Footnote 10}

A total of 40 tables were produced for the 2001 Census. Geographically, 28 tables were produced at the national level, 9 were produced at the 'COROP' (regional) level and 3 were produced at the municipal level. In terms of content, 8 tables covered housing, 2 covered commuting and 30 covered demographics (occupation, education, economic activity). Sub-city tables were produced for 10 large cities.

Statistics Netherlands has stated that the total cost of the 2001 Virtual Census was approximately 3 million euros (excluding the costs of the Population Register, which already existed), compared to an estimated cost of 300 million euros to conduct a traditional census. The approach also permitted faster production of results than those produced by other European countries using the traditional census approach.

Statistics Netherlands' plan for the 2011 Census is similar to that used in 2001. The SSD has continued to develop through the addition of sources such as a new Housing Register and the merger of other registers, and is now used to support a wide range of statistical production. Dutch legislation passed in November 2003 also provides Statistics Netherlands with increased authority to access administrative sources, at no charge. As well as using the SSD in the 2011 Census, the plan is to base all household statistics on the SSD by 2013. As well, the SSD serves as a valuable resource for methodological purposes, such as adjustment for survey non-response, which is relatively high in the Netherlands.

New European census legislation passed in 2008 mandates the production of census data on specific topics and at specific levels of detail that go beyond what was produced in 2001. As of 2009, Statistics Netherlands was examining whether the repeated weighting method will be able to meet these additional requirements in 2011.

Germany

City-level censuses in what is now Germany were held as early as 1471. Regular periodic censuses were carried out in the German Empire, first in 1871 and then every five years from 1875 to 1910. During and between the two world wars, censuses were held at irregular intervals, varying between one and eight years. In the postwar period, East Germany carried out censuses in 1950, 1964, 1971 and 1981, and West Germany held censuses in 1950, 1956, 1961, 1970 and 1987.

The last West German census was originally planned for 1980, but was postponed to 1983 and then to 1987 due to widespread public objections. As a result of legal cases brought against the government, changes to several aspects of the census were made by the Constitutional Court, and the modified census was eventually carried out in 1987. A concurrent census in East and West Germany was planned for 1991, but was cancelled with the fall of the Berlin Wall and reunification.

Since the last censuses in 1981 and 1987, official population figures have been based on intercensal population updates. However, the Federal Statistical Office believes that the official population of Germany is overestimated by about 1.3 million persons (out of approximately 80 million) because the basic data on which these estimates are based are seriously out of date (Federal Statistical Office 2011). As well, because the new EU census legislation requires that all member countries take censuses of population and housing every 10 years, Germany is conducting a 2011 Census, with a reference date of May 9.

Following methodological testing in 2001, the governing parties of the coalition government agreed in November 2005 to proceed with a 2011 Census, and on August 29, 2006, the federal cabinet decided that the 2011 Census would be register-based. On December 13, 2007 a census preparation law was enacted to enable preparations for the census to proceed, and on December 3, 2008 the 2011 Census Law was enacted to provide the legal basis for conducting the 2011 Census. The law covers general policy, collection and consolidation of data, generation of households, organization, measures to assure the quality of the census results, the obligation to give information, data protection and final provisions (Gerull 2009).

The 2011 Census combines administrative registers, complete enumeration and sample surveys. The two main administrative sources are the municipal population registers, which contain comparable information from approximately 12,600 municipalities, and register information from the Federal Employment Agency. The latter contains information on employees subject to compulsory social insurance contributions, on persons who are registered as unemployed or looking for work, and on civil servants, army personnel and judges. Two versions of the municipal registers are obtained, one as of the census date and a second version at a later date with updates applied by the municipalities. Because there is no central population register and persons are sometimes registered in more than one municipality, a large matching operation is undertaken to attempt to eliminate duplicate registrations.

The administrative registers have several weaknesses for census-taking purposes that require them to be supplemented by both complete enumerations and sample surveys (Siedt 2006). First, coverage of the municipal population registers is known to be problematic for residents of institutions and collective dwellings. As well, the rates of undercoverage and overcoverage of the municipal population registers vary significantly by municipality size. For the registers from the Federal Employment Agency, some persons (e.g., the self-employed) are not well covered. Finally, some of the variables required by EU census legislation are not available in the registers, for example, details of employment, migration and education and training.

The first complete enumeration is the Census of Buildings and Housing, conducted by mailing a questionnaire to all owners (not residents) of residential property. The enumeration is based on a nation-wide register of all buildings established specifically for the 2011 Census by the Federal Statistical Office and the statistical offices of the Länder, using sources such as files of land surveying offices, residents' registration offices and the Federal Employment Agency. The Census of Buildings and Housing questionnaire collects information such as the year of construction, type of building, equipment (e.g., bath or shower), floor space and number of rooms.

A complete enumeration is also conducted at residential establishments and collective living quarters, with the primary objective of addressing the weakness of the municipal population registers in covering these populations. For 'non-sensitive' residences (e.g., student residences, senior citizens' or care homes) the residents are interviewed in person. In 'sensitive' facilities (e.g., psychiatric hospitals and refugee establishments) only basic information, sufficient to establish the number of persons, is collected from the facility management.

The sample survey component will be conducted by personal interviews with a 10% sample of the population. Persons in non-sensitive residences are eligible to be included in the sample survey and the survey data will be collected at the same time as the enumeration of residential establishments and collective living quarters. The results of the survey will serve two purposes. First, they will be used to statistically adjust^{Footnote 11} the municipal register counts for undercoverage and overcoverage to produce official population estimates. Second, the survey will produce data for required variables not available in the registers. A further 5% of the sample survey will be interviewed in a follow-up survey a few weeks later to measure the quality of the main sample survey.

In order to ensure high-quality data, responses to the complete enumerations and the sample surveys are compulsory. The final census results will be produced by linking these sources of data (i.e., registers, complete enumerations and sample surveys) together at the micro-data level. For example, household relationships will be established by linking the various components in a 'household generating procedure,' allowing the production of estimates such as the average floor space of families with three or more children.

The new German census approach of combining registers, complete enumerations and sample surveys is methodologically complex, but it is expected to be less expensive and to have reduced respondent burden compared to a traditional census. As this is the first time such an approach has been used in Germany, it is not yet known how successful it will be.

Switzerland

Switzerland has carried out a census almost every 10 years from 1850 to 2000. While previous Swiss censuses used administrative registers in a subsidiary role, Switzerland is making major changes for its 2010 Census (Federal Statistical Office 2008). The necessary legislation was adopted in June 2007, with an effective date of January 2008. The reference date of the 2010 Census was December 31, 2010.

The 2010 Swiss Census consists of four components.

The first component is an annual register survey that brings together data from the various municipal and canton-level population registers and data from the Federal Register of Buildings and Dwellings. The register survey will produce basic population, household and housing statistics on a census basis. To facilitate this new register survey, the new legislation mandated the following changes:

• harmonization of the municipal and canton-level population registers, in terms of their core variables, format, standards for completeness and accuracy, and protocols for electronic data exchange
• replacement of the existing social security number by a new 13-digit PIN issued to every Swiss resident and inclusion of the PIN on all registers
• assigning to each person in the population registers the nine-digit federal building identification number and, in addition, the three-digit dwelling identification number, so that household composition can be determined from the register
• linkage of persons in sample surveys to the person register by means of the PIN
• linkage of persons to the Business Register in order to link them to their employers.

The second component is an annual 'structural' survey of 200,000 persons of age 15 and over, or about 3.5% of this population. The content is similar to that of a typical census (e.g., employment, mobility, education, language, religion and culture, and family structure). Response to the annual structural survey is mandatory. Data from the survey can be pooled over several years to form time-period average estimates for small areas and subgroups, similar to the U.S. American Community Survey. Cantons and cities have the option to buy additional sample for their areas.

The third component is a series of topic-based telephone sample surveys of 10,000 to 40,000 persons, rotating among five topics (mobility, education, health, families, and language, religion and culture) over a five-year period. These surveys will provide data at the national level and for the seven regions of Switzerland.

The fourth component is an annual omnibus survey of 3,000 persons which will provide rapid answers to current questions. Results will be at the national level only.

The new 2010 Census is expected to provide more frequent and more highly integrated data at a lower cost, with reduced burden on municipalities, cantons and the population.

Austria

Austria is unique in moving from a completely traditional census in 2001 to a completely register-based census for 2011 (Lenk 2008, Fiedler et al. 2010). The legal basis for the new approach is the Register-based Census Act, passed in March 2006. A register-based census test was held in 2006 and the full census will be held in 2011.

The 2011 Austrian Census will be conducted by linking together eight existing 'base registers': the Housing Register of buildings and dwellings, the Central Population Register, the Register of Educational Attainment, the Register of enrolled pupils and students, the Central Social Security Register, the Tax Register, the Unemployment Register, and the Business Register of enterprises and their local units.

Unlike other countries that conduct a register-based census, the Austrian Data Protection Commission (DPC) does not permit the registers to be linked by a common PIN. Instead, the registers are linked by an artificial identifier, known as the 'branch-specific personal identification number for official statistics' (bPIN OS). This identifier is generated by the DPC from the PIN of the individual held on the Central Population Register, using a complex algorithm known only to the DPC. Statistics Austria then supplies the bPIN OS to the owners of the other registers in encrypted form, who return the data together with the encrypted bPIN OS and a number of their own. The latter can be used in case of follow-up questions by Statistics Austria.

One of the basic principles is that of redundancy. Because these registers were not connected in the past and each register collects its data independently, there are often data inconsistencies across registers that must be resolved. By matching a large number of registers, Statistics Austria hopes to determine the most 'plausible' value in the case of data inconsistencies. As well, to improve and verify the quality of the census results, the eight base registers will be linked to seven other 'comparison registers': the Child allowance register, the Central foreigner register, Registers of public servants of the federal state and the Länder, the Register of car owners, the Register of social welfare recipients, the Conscription Register, and the Register of alternative civilian service. The Register-based Census Act specifies which registers are base registers and which are comparison registers.

One of the main limitations of the new approach is that some of the data that were collected in previous censuses will no longer be available. In particular, the registers do not contain information on languages spoken, religion, mode and duration of commuting, and, most importantly, occupation. However, there have been some changes to the registers, such as the addition of the place of work to the social security records, to provide annual commuting data.

United States

The first U.S. Decennial Census was conducted in 1790. Sampling was introduced in the 1940 Census, when a long form was completed by 20% of households, with the rest receiving a shorter form than in previous censuses. In the 2000 Census, the long form was sent to approximately one in six households. For the 2010 Census, the long form was replaced by a large-scale continuous survey known as the American Community Survey (ACS).

The origins of the ACS date back to the early 1980s (U.S. Census Bureau 2009). In response to user demand for more timely data than the Decennial Census provided, continuous measurement was proposed as an alternative (Kish 1981). The U.S. Congress also recognized the need for more frequent data and authorized a mid-decade census for 1985, but the funds to conduct it were never allocated.

The idea of continuous measurement was resurrected again in the early 1990s, and in 1993, the U.S. Census Bureau developed a set of research proposals for an ACS. Three possible prototypes were outlined, and in 1994, one of the prototypes was chosen for further development. The survey would involve collection by mail, followed by telephone and then personal visit follow-up for non-response. Because the ACS was viewed as a replacement for the Decennial Census long form, response would be mandatory. In November 1995, testing of the ACS began in four counties, and was expanded to a further five counties in November 1996. Three additional counties were added in 1998, and in 1999, the test was expanded to 36 counties in 26 states. In 1999 and 2001, the test was expanded to cover Group Quarters (known as collective dwellings in Canada).

The primary purpose of the tests in the 1990s was to refine the methodology. Based on their success, it was decided to proceed, and in 2000, a large-scale demonstration ACS was introduced. The demonstration ACS included the original 36 counties and added 1,203 new ones (although the latter used an area sample design similar to that of the Current Population Survey) and had a total sample size of 866,000 households. The demonstration ACS had two purposes: first, to evaluate the feasibility of such a large-scale continuous survey from a methodological and operational perspective, and second, to compare the quality of the data from the demonstration ACS to that of the 2000 Census long form. The demonstration ACS was successful on both counts, and plans were developed for full implementation.

In 2003, at the request of the Congress, the Census Bureau conducted a test of making response to the ACS voluntary (U.S. Census Bureau 2003). The voluntary ACS resulted in more than a 20 percentage point drop in the mail cooperation rate and a reduction in the reliability of the data. The test also estimated that a voluntary approach would cost at least an additional $59.2 million annually due to the increased telephone and personal visit non-response follow-up needed to obtain results of similar quality to a mandatory survey. The ACS has continued to be a mandatory survey.

The final phase of the ACS was its full implementation. Implementation was originally planned for 2003, but was delayed for two years due to budget restrictions. In January 2005, full implementation began in all 3,141 counties of the United States, with a sample size of three million households each year. After subsampling for non-response follow-up, responses are received from approximately two million households each year. Group Quarters were added to the full ACS in January 2006.

The first set of ACS one-year estimates was produced in 2006, three-year estimates were first produced in 2008, and five-year estimates were first produced in December 2010. The ACS estimates are calibrated to the intercensal population estimates at detailed levels of geography, rather than to the census counts as the long-form data were previously.

To date, the ACS has been considered a success, although a number of challenges remain for both the Census Bureau and data users (for example, see National Academies Press 2007). Throughout the development, testing, demonstration and implementation phases, the Census Bureau has worked closely with its data users, advisory groups, and other stakeholders to ensure that the new approach is understood.

The 2010 Decennial Census included only the basic demographic, race, Hispanic origin and housing questions, as previously asked on the short form. The Census Bureau's planning for the 2020 Census assumes that it will continue to be a short-form census. The main driver for planning the 2020 Census is the rising cost of the traditional census, which it attributes to declining self-response rates, paper-based and labour-intensive methods requiring a large field infrastructure, and substantial investments in updating the address frame (Weinberg 2010, Groves and Vitrano 2011). The Census Bureau has solicited ideas from a number of sources, including a panel convened by the National Academy of Sciences (National Academies Press 2010) and Statistics Canada; it has consolidated these ideas into some 75 possible research projects, and it has prepared a budget initiative for work to begin in October 2011. Current design alternatives range from simply enhancing the baseline census conducted in 2010 to an administrative record-based enumeration with field follow-up to fill data gaps. The Census Bureau will continue to seek input from stakeholders and the National Academy of Sciences as its plans for 2020 evolve.

France

Napoléon Bonaparte established the first traditional census in France in 1801. From then until the Second World War, the French census was conducted every five years, with a few exceptions. Following the Second World War, the census was conducted at irregular intervals: 1946, 1954, 1962, 1968, 1975, 1982, 1990 and 1999. The irregularity was in part because there was no legal requirement for a census to be taken in specific years.

In 1997, the Institut national de la statistique et des études économiques (INSEE) began looking at alternatives to the traditional census (Clanché 2010). One of the main drivers was the difficulty in securing budget for the census; the 1999 Census was originally supposed to have been conducted in 1997 but was delayed for budgetary reasons. Other factors were the increasing demands from users, particularly at the local level, for more regular and timely data, the difficulty in negotiating arrangements for the census with the local authorities ('communes') and the difficulties in controlling large-scale census operations involving the collection of data from 20 million households over a four-week period.

In February 2002, legislation was passed that paved the way for France to move to a rolling census for establishing, on an annual basis, the legal populations of its various administrative units, from the commune level up to the regional and national levels. After a period of user consultation, methodological development and testing, the rolling census was introduced in 2004 and its first set of official estimates based on five years of data collection (2004 to 2008) became available in December 2008. The rolling census is conducted in two different ways, depending on the size of the commune in which it is conducted.^{Footnote 12} Response to the census is required by law.

For communes with less than 10,000 residents, a complete census is conducted once every five years on a rotating basis. For example, a commune may receive the census in 2004, 2009, 2014, etc., while a neighbouring commune may receive the census in 2005, 2010, 2015, etc. The five panels of communes thus formed are 'balanced' on a number of characteristics from the 1999 Census so that each panel is as representative as possible at the region level.

For communes with a population of greater than 10,000 persons, an 8% sample of addresses is surveyed each year. The samples are chosen from an address register ("Répertoire d'immeubles localisés") established from the 1999 Census, and kept up to date from administrative sources and with assistance from the local authorities. Thus, over a five-year period, some 40% of addresses are surveyed.

The 'legal population' for each geographic area is established annually, with a reference date of January 1 of the middle year of the previous five-year period. For example, the first set of estimates based on the 2004 to 2008 collection had a reference date of January 1, 2006. Because data collection is spread over the five year period and is conducted differently in large and small communes, complex mathematical adjustments, involving interpolation and calibration, are required to bring the data to the common reference date. For the large communes, the official population figures are established on the basis of a 40% sample, not a complete enumeration.^{Footnote 13}

Detailed socioeconomic data are disseminated a few months after the population figures. These data are also based on the previous five years of data collection, and are adjusted in order to maintain coherence with the population figures.

Responsibility for statistical data collection is much more decentralized in France than in Canada. While INSEE is responsible for establishing the protocols to be followed and the questionnaires to be used, and is the sole recipient of the confidential data, it is the local communes who are responsible for the actual data collection, i.e., hiring, supervising and paying enumerators, organizing field operations, etc. Questionnaires are dropped off by enumerators and collected by them on a return visit.