Census-taking approaches

This section describes each of the census-taking approaches and the necessary conditions for its use. The main strengths and weaknesses of each approach are also described, taking into account factors such as adherence to the five essential UN/UNECE features, costs, response burden, implications for census content, frequency and timeliness of data, data accuracy, and risks particular to the approach.^{Footnote 1}

Traditional census

Description

With the traditional census approach, census population and housing data pertaining to a specific point in time (i.e., a 'Census Day') are obtained from a complete enumeration^{Footnote 2} of the population and housing units. The data may be collected by enumerators or through self-completion of a questionnaire by the household. Various collection modes may be used, such as personal visits, mail, telephone and Internet. One questionnaire may be used to collect the data from all households, or a long questionnaire may be completed by a sample (generally very large) of households, with the remainder of households receiving only a short questionnaire. The traditional census is ideally conducted at regular intervals, such as every 5 or 10 years, to facilitate comparisons over time. The traditional census is the approach used by the majority of countries in the world and in the UNECE region.

Necessary conditions

For a traditional census to be successfully used, the following conditions must exist:

• There must be a high level of awareness and cooperation by the public to participate in the census.
• There must be a geographic infrastructure for the census, such as a set of maps (if questionnaires are distributed by or completed by enumerators) and/or a list of addresses (if questionnaires are mailed out), that permits the geocoding of all questionnaires to very small geographic areas.
• There must be a stable political and socially secure environment in the country, both to encourage public participation and to provide safe conditions for enumerators.
• If self-completion is used, there must be a relatively high level of literacy in the population.
• The country must have the legal and administrative framework and the resources, both human and financial, to undertake a traditional census.

Strengths

The traditional census has the following main strengths:

• It fulfils all five of the UN/UNECE features of a census: individual enumeration, universality, simultaneity, defined periodicity and production of small-area data.
• From a data user perspective, the traditional census provides a snapshot of the entire country at one point in time and provides data for small areas and subgroups of the population that may be available from no other source.
• The choice of content of the traditional census is relatively flexible. It is normally based on a balance of the statistical needs of users, costs, data quality and the intrusiveness/burden on respondents.
• Widespread public participation generates interest in seeing the results, contributing to the usefulness of the census and statistics more generally.
• The traditional census can be used to identify small subgroups of the population to be followed up with more detailed post-censal surveys.

Weaknesses

The traditional census has the following weaknesses:

• Because it requires the individual enumeration of every housing unit and person in the country, a traditional census is relatively expensive.
• The costs and complexity of a traditional census mean that it can only be conducted every 5 or 10 years. As a result, some of the data may become out of date before the next census is conducted.
• The large volume of data to be collected, captured and processed usually means that several months elapse from Census Day to the dissemination of the results.
• The traditional census imposes some amount of burden on the population. There may also be duplication of reporting if some of the information has already been reported to the government.
• Because a high level of public participation is required, with the associated high visibility, the traditional census is subject to changing public and government attitudes about the trade-offs among data requirements, quality, costs and intrusiveness/burden issues. It can also become the focus of protests that have little to do with the data collection itself.
• Funding is concentrated in the one- or two-year period around the census year rather than being spread evenly.
• The traditional census is subject to unforeseen events (e.g., natural disasters, labour disruptions and elections) that can seriously affect the census and, in some cases, limit the possibility to recover.

Census employing existing administrative registers

Description

With the register-based approach, census data are produced by linking together existing administrative data sources such as population registers, business registers, building/dwelling registers, employment registers, education registers and taxation registers. To date, such register-based censuses have been conducted primarily in the A7, although an increasing number of countries (mostly in Europe, see Survey of census methods for the 2010 round of censuses) are moving to the use of registers for conducting at least a portion of their censuses in the 2010 round. A few countries base their entire census on registers, while other countries generate some of the data from registers (e.g., population counts and basic demographic data) and produce the rest of the required data from a complete enumeration of the population and/or sample surveys. Because all such variants use administrative registers as the basis of the census, they have similar necessary conditions, strengths and weaknesses.

Necessary conditions

Several conditions are necessary for the adoption of a register-based approach (Tønder 2008, UNECE 2007, and Redfern 1989).

• There must be a legal basis giving the statistical agency the right to access administrative data at the unit level and to use identification numbers to link various administrative data sources together for statistical purposes. The legislation must also provide a detailed definition of data protection; for example, it should specify that the statistical data produced by the linkage process may not be fed back to the administrative databases (known as the 'one-way traffic' principle.)
• There must be public understanding and approval of the use of administrative data for statistical purposes and public recognition of the advantages of using data already collected for administrative purposes compared to collecting the data again.
• There must be a universal Personal Identification Number (PIN) that can be used to link administrative data across sources at the unit level. In the Nordic countries where register-based censuses are most developed, the PIN is used for virtually all transactions with the government, at both the national and local levels (e.g., driver's license, employment, school enrolment and health care).
• The country must have a well-developed set of register systems that fulfil administrative needs but that also contain data covering the most important subject areas for the statistical system. At a minimum there must be a population register, a business register, and a building/dwelling register. The coverage of these registers and the quality of the data contained within them must be sufficiently high to be useful for statistical purposes.
• There must be incentives, such as a legal requirement, for the population to register and to inform the register authorities of changes of address. There must also be a reliable way to assign the units included in the registers (e.g., persons, businesses, dwellings) to a detailed geographic level (geocoding) in order to produce small-area detail.
• Dates of changes or events (e.g., change of address, change of employers, births, deaths, immigration and emigration) must be reliably recorded with minimal delay in recording the event.

Strengths

The register-based approach has the following main strengths:

• Like the traditional census, the register-based census can fulfil all five of the UN/UNECE features of a census: individual enumeration, universality, simultaneity, defined periodicity and the production of data for small areas and subgroups, although the extent to which it does so may depend on the specific registers that are used.
• Like the traditional census, the register-based census can provide a snapshot of the entire country at one point in time and can provide data for small areas and subgroups of the population.
• Because it uses data already collected for administrative purposes, the marginal costs to produce census data from registers are generally much less than for a traditional census.^{Footnote 3}
• A register-based census can generally produce data on a yearly basis, compared to every 5 or 10 years for a traditional census.
• The data may be available just a few months after the reference date (the exact timeliness will depend on how quickly the registers are updated).
• A register-based census reduces the burden on respondents, because it reuses data that have already been collected for other purposes.
• Funding can be more evenly spread than for the traditional census conducted only every 5 or 10 years.
• A register-based census is less susceptible to unforeseen events, since it is based on administrative registers instead of large field collection operations.
• The register-based census may be used to identify small subgroups for post-censal surveys, depending on the variables contained in the registers.

Weaknesses

The register-based census has the following main weaknesses:

• The census content is limited to the data variables already available in the administrative registers. This restriction can be eased by combining a register-based census with a complete enumeration and/or sample surveys (either ad hoc or pre-existing) to produce content that is not available in the administrative registers, although at increased costs and respondent burden.
• The concepts and definitions that apply to data in the registers may not correspond to those desired for statistical purposes.
• Linkage of different administrative registers for the same unit may result in data inconsistencies that need to be resolved.
• The register-based census is subject to changes in legislation and administrative practices on the part of the register authorities.
• Development of a register-based census is generally a long process, taking several years or even decades, with a significant front-end investment, and possibly requiring negotiated agreements across multiple levels of government.

Traditional census with yearly updates of characteristics

Description

As the name suggests, the traditional census with yearly updates of characteristics consists of two parts. A traditional census is conducted periodically to collect basic information (e.g., population and dwelling counts, age and sex), and a large-scale sample survey collects the more detailed population and housing information on a continuous basis. To date the only country using this method is the United States.^{Footnote 4} The American Community Survey consists of a sample of 250,000 households each month and asks the questions that were previously on the long form of the Decennial Census. As with the traditional census, various modes of data collection are possible for both the periodic census and the continuous survey.

Because data collection for the survey portion is continuous rather than at a single point in time, the data must be pooled over time to produce estimates. In the case of the American Community Survey, the estimates represent averages for one-, three- and five-year periods. The length of the time period depends on the size of the geographic area; estimates for large areas such as states or large cities may be one-year averages, while for small municipalities, counties or census tracts they may be three- or five-year averages. Because the estimates do not refer to a fixed point in time, the interpretation and analysis of these time-period averages can be somewhat complex. Once the survey is established, however, updated results can be produced each year for all geographic areas by replacing the earliest year of data with the most recent year.

Necessary conditions

The traditional census with yearly updates of characteristics approach requires the following conditions to be met:

• It generally requires a multi-year program of comprehensive planning, development and testing to implement.
• It requires the agreement of census stakeholders and policy-makers to move from a once-every-5- or 10-year snapshot to an annually updated multi-year average approach.
• A geographic infrastructure (address list or maps) is required, but unlike the traditional census, the infrastructure must be continuously updated rather than being updated once just before the census.
• Several years of data collection (three or five) are required before the first data for smaller areas can be made available.
• It requires substantial annual funding rather than funding clustered in a one- or two-year period around census year.
• Because it is more complex, both in terms of its operations and the resulting data, it requires a high level of professional staff to maintain it and users with the ability to interpret the data.

Strengths

The traditional census with yearly updates of characteristics approach has the following strengths:

• The traditional census part of the approach meets all of the UN and UNECE features of a census, although only for the basic data.
• From a data user perspective, the traditional census part provides a snapshot of the entire country and for small areas and subgroups of the population, although only for the basic data.
• The content has the same degree of flexibility as a traditional census and, in theory, can incorporate new data needs relatively quickly into the continuous survey part, although at the expense of temporarily reduced comparability for smaller geographic areas that rely on aggregated averages over a period of three to five years.
• Once established, the continuous survey part of the method produces updated data every year.
• Funding is more evenly spread out over the census period rather than being concentrated.
• The smaller scale of the continuous survey coupled with a more permanent field staff permits tighter control over non-sampling errors.
• Improvements to the methodology of the continuous survey can, in theory, be made relatively quickly, although at the expense of temporarily reduced comparability for smaller geographic areas that rely on aggregated averages over a period of three to five years.
• The reduction of the traditional census part of the method to a basic census allows concentration of efforts on the most important aspects of the census (e.g., population counts).
• The continuous survey part of the method reduces the operational risks, as there is an opportunity to recover from unforeseen events.

Weaknesses

The main weaknesses of the traditional census with yearly updates of characteristics approach are:

• The continuous survey part of the approach does not respect the UN's simultaneity feature of a census because data must be accumulated over time. The resulting data represent time-period averages rather than a point-in-time 'snapshot,' and may thus be more difficult to interpret and to compare to other data.
• Both the traditional census and the continuous survey impose some amount of burden on the population. There is also a small amount of duplication at the time the traditional census is conducted because respondents in the continuous survey are also asked to respond to some of the same questions on the census. There may also be duplication of reporting if some of the information has already been reported to the government.
• Although new estimates can be produced annually, multi-year averages do not represent completely new data. For smaller areas the 'average age' of the data may be substantial. For example, the United States published its first five-year estimates in December 2010, based on data collected from January 2005 to December 2009. From the average date of the data collection period (i.e., July 2007) to the publication date (December 2010) is approximately three and a half years.
• Compared to the data from a long questionnaire collected as part of a traditional census, where the short-form data can be used to calibrate the long-form data, the precision of the continuous sample survey will be lower for the same sample size (the sampling error will be higher).

Rolling census

Description

The rolling census also uses a continuous sample survey, but instead of conducting a periodic traditional census, the continuous sample survey is designed so that it covers 100% of the country over some time period (e.g., five years). To date, the only country using a rolling census is France.^{Footnote 5} Many of the necessary conditions, strengths and weaknesses are similar to those of the traditional census with yearly updates of characteristics. As with the traditional census conducted at a point in time, a variety of collection methods can be used, including long and short forms, canvasser or self-enumeration methods, and various response modes.

Because data collection is continuous rather than at a single point in time, data from the rolling census must be pooled over time. In the case of the French rolling census, complex mathematical methods involving calibration and interpolation are used to transform data collected over a five-year period into estimates with a single reference date. However, once the rolling census is established, updated estimates can be produced each year by replacing the earliest year of data with the most recent year.

Necessary Conditions

The rolling census approach requires the following conditions to be met:

• It generally requires a multi-year program of comprehensive planning, development and testing to implement.
• It requires the agreement of census stakeholders and policy-makers to move from a once-every-5- or 10-year snapshot to an annually updated multi-year approach.
• A geographic infrastructure (address list or maps) is required, but unlike the traditional census, the infrastructure must be continuously updated rather than being updated just once before the census.
• Several years of data collection (five in the case of France) are required before the first data for smaller areas can be made available.
• It requires substantial annual funding rather than funding clustered in a one- or two-year period around census year.
• Because it is more complex, both in terms of its operations and the resulting data, it requires a high level of professional staff to maintain it and users with the ability to interpret the data.

Strengths

The rolling census approach has the following strengths:

• The content has the same degree of flexibility as a traditional census. It is also quite flexible for being able to incorporate new data needs relatively quickly, though at the expense of temporarily reducing comparability of the data.
• Once established, the rolling census produces updated data every year.
• Funding is evenly spread out over the census period rather than being concentrated.
• Improvements to the methodology of the rolling census can, in theory, be made more frequently than once a decade, although at the potential expense of temporarily reducing comparability of the data.
• The rolling census approach reduces the operational risks of the census, as there is an opportunity to recover from unforeseen events.
• The smaller scale of the rolling census permits tighter control over non-sampling errors.

Weaknesses

The rolling census has the following weaknesses:

• In order for the rolling census approach to respect the UN/UNECE simultaneity feature of a census, complex estimation methods are needed. In the French implementation, mathematical techniques are used to transform data collected over a five-year period to a single reference date,^{Footnote 6} and sampling is used to produce the official population estimates.^{Footnote 7} The resulting estimates are not simply the results of an enumeration but incorporate a degree of modelling and sampling error.
• The rolling census imposes some amount of burden on the population. There may also be duplication of reporting if some of the information has already been reported to the government.
• Because of the rolling approach, persons who move may be enumerated more than once or may be missed. The effects of this may not cancel out, particularly for small areas.
• Although data are produced more frequently (annually), the interval from the reference date of the data to the publication date may still be substantial. For example, France published its first five-year estimates in December 2008 with a reference date of January 1, 2006, a lag of three years.