Assessment of census approaches in the Canadian context

Canada:

Overall, public cooperation with the Canadian census has been excellent. For example, the response rate to the 2006 Census long-form questionnaire was 94%. While non-response rates have risen slightly during the past few censuses, they do not appear to have reached levels which bring into question the quality of the data.

In mid-2010, the government, citing privacy reasons, decided that the 2011 Census would consist of 10 questions, and asked Statistics Canada to provide options, from which the government chose one, for collecting the remaining proposed questions on a voluntary basis. The level of public cooperation with the voluntary NHS is as yet unknown but is expected to be lower than with the 2006 approach, when response was mandatory.^{Footnote 1}

---

Other countries:

Several countries have cited declining participation rates as a concern (e.g., ONS 2003a). In some European countries, response rates to surveys and censuses have fallen to levels much lower than in Canada, to the point where the traditional census had to be cancelled. For example, a pilot test for the cancelled 1981 Dutch Census estimated that the non-response rate would be 26% (Bethlehem 2010).

Canada:

The Address Register (AR) was used to mail out invitation letters or questionnaires to approximately 80% of households in 2011. In the remainder of the country, questionnaires were dropped off by enumerators (18%), or the census was completed by the canvasser method (2%), using a well-developed set of maps. Coverage of the AR has increased since its first use for mail-out in the 2006 Census.

---

Other countries:

All countries that conduct a traditional census have geographic infrastructures consisting either of address lists, maps, or a combination of the two. (United Nations 2011b, Chapter 3).

Canada:

Canada has a very stable political and socially secure environment that supports both public participation and enumerator safety. There has never been an occasion where the census has had to be cancelled or postponed due to political instability or enumerator safety concerns.

---

Other countries:

Most developed countries are similar to Canada in terms of having conducted censuses on a regular basis. In less developed countries, particularly in Africa, censuses have occasionally been postponed for reasons of political instability (United Nations 2011b).

Canada:

Starting with the 1971 Census, self-completion has been the primary mode of enumeration in Canada. Compared to the canvasser (face-to-face) methodology, self-enumeration has important advantages in terms of privacy, costs and data accuracy. The addition of the Internet as a response option for the 2006 Census furthered these advantages.

For population groups where barriers of language or literacy do exist, the canvasser method is still used. The census questionnaire is translated into 20 ethnic and 11 aboriginal languages, as well as being available in braille, audio and signed video.

---

Other countries:

Self-enumeration tends to be used in developed countries for the reasons cited above, while face-to-face interviews are more common in less developed countries (e.g., India). Globally, the most common mode of enumeration is face-to-face interview, used by 85% of countries. Self-enumeration by paper questionnaires is the next most common mode (28% of countries), followed by Internet-based self-enumeration at 23% (United Nations 2011b, Table 4.1).^{Footnote 2}

Canada:

The requirement to conduct the decennial census in years ending in '1' is part of the Constitution Act and dates to 1870, the earliest of any country in the UNECE (UNECE 2008, Table 4.1). The quinquennial census, conducted in years ending in '6', is required by the Statistics Act but is also a constitutional requirement, with the original obligation for quinquennial censuses recorded in the three acts that brought Manitoba, Saskatchewan and Alberta into Confederation.

Statistics Canada is charged with conducting the census and is allocated the financial resources to do so. Compared to other countries conducting a traditional census, particularly the United States, the costs per household in Canada have remained relatively stable over the past several censuses.

---

Other countries:

Most countries have either a census act or a statistics act as the legal basis for census-taking. For the 2000 census round in the UNECE region, all countries had one or the other and most countries had both (UNECE 2008, Table 4.1).

Globally, it is expected that 99% of the world's population will be counted in the 2010 census round.^{Footnote 3} The UN encourages all countries to conduct a census and provides extensive training and other forms of support (United Nations 2011a).

Canada:

Section 13 of the Statistics Act gives Statistics Canada the right to access records held in any department or municipal office, corporation, business or organization for the purposes of the Act. However, in practice, Statistics Canada negotiates access to administrative data with each organization on a case-by-case basis. In many cases, other organizations are subject to laws that prohibit or limit the sharing of data under their control.

There is no explicit mention of record linkage in the Statistics Act itself. Statistics Canada's Policy on record linkage (Statistics Canada Policy Manual, 4.1) states that record linkage will only be undertaken when the public good clearly outweighs the potential invasion of privacy. All cases must be approved by the Chief Statistician and are posted on Statistics Canada's website.

In a situation where administrative registers would be combined with complete enumeration or survey collection, Statistics Canada's Policy on informing survey respondents (Statistics Canada Policy Manual, 1.1) would require that respondents be notified at the time of survey collection of any planned linkages of their survey responses to other data files, where linkage is for other than internal methodological purposes.

---

Other countries:

New census legislation may need to be introduced (e.g., Sweden, Netherlands, Switzerland, Austria) for countries to adopt a register-based approach involving access to and linkage of administrative registers. In the case of Sweden, more than 10 years was required for such legislation to become law; as a result, Sweden did not conduct a census in the 2000 round. In other cases (e.g., Switzerland, Austria), it was possible to adopt new legislation and put the new approach in place within a 10-year period.

All countries that conducted their census in the 2000 round using administrative registers (either combined or register-only) had data protection laws in addition to their census act or statistics act (UNECE 2008, Table 4.1).

Canada:

There appears to be little or no research into Canadian attitudes toward the possibility of conducting a census using administrative registers. A partial indication may come from the 2006 Census, which gave respondents the option to allow Statistics Canada to obtain their income data from tax files, rather than completing the income questions on the census questionnaire. The national-level take-up rate for this option was approximately 82%.^{Footnote 4}

In contrast was the public reaction in 2000 to the Longitudinal Labour Force File, created by Human Resources Development Canada (HRDC) in the 1990s as a research database intended to improve overall program efficiency. The file linked together personal tax records, child tax benefit records, the immigration and visitors file, provincial and municipal welfare files, and several others. The file was criticized by the Privacy Commissioner in his 1999-2000 Annual Report to Parliament (Privacy Commissioner 2000) as a 'de facto citizen profile.' The subsequent negative public reaction led the government to destroy the file. However, it should be noted that the file was intended for research purposes rather than statistical purposes, and some of the Privacy Commissioner's concerns were about the lack of legal protection of the information in the file and the lack of transparency about the existence of the file.

Overall, the public acceptability of a register-based census is largely an unknown quantity, but it could reasonably be expected to be opposed by at least some members of the public.

---

Other countries:

The acceptance and use of population registers for administrative purposes is deep-rooted in many European countries, often dating back several decades or even centuries (Redfern 1989). Where such population registers already existed, the transition to using them for statistical purposes appears to have been a logical progression for both the public and the government.

In other countries, attempts to introduce population registration have been unsuccessful. As noted above, the attempt to introduce a National Identity Card in the United Kingdom met with considerable public opposition and was scrapped in early 2011. In Australia, attempts to introduce a universal 'Australia Card' in the mid-1980s and a Health and Social Services Access Card (dubbed 'Australia Card Lite') in 2006 both met with strong public opposition and were abandoned (Parliament of Australia 2006, Redfern 1989).

Canada:

Canada has no universal PIN. The closest to this is the Social Insurance Number (SIN), but it is not universal and its usage is strictly limited by legislation and Treasury Board of Canada Secretariat policy.^{Footnote 5} The use and administration of the SIN has been the subject of numerous studies and audits over the years.^{Footnote 6} On each such occasion since the introduction of the SIN in 1964, the government has rejected its use as a personal identifier.

A wide variety of identification numbers are used for other administrative purposes (e.g., health insurance, drivers' licences, education, passport). There are no links between these other identifiers and the SIN.

---

Other countries:

All of the countries that conduct a register-based census have universal PINs that are used to link registers at the unit level. In many countries, these PINs are used to access a wide variety of government services. In the case of Switzerland, a new PIN was issued to all residents as part of the new census legislation. In other cases (e.g., Austria and the Netherlands), special approaches involving encryption of PINs may be needed to use them for statistical purposes.

Canada:

Canada has no population register analogous to those in countries that conduct a register-based census. Although there are a number of administrative databases (e.g., income tax, electoral rolls, and provincial/territorial health care records) that cover parts of the population for specific administrative purposes, none of these registers is sufficiently comprehensive for census purposes, nor are there common identifiers that would permit them to be linked to create a more comprehensive database. Several of these administrative registers (e.g., health insurance, drivers' licences) are under provincial/territorial jurisdiction.

The possibility of developing a national identity card system for Canada was raised by the minister of Citizenship and Immigration in 2003. The Standing Committee on Citizenship and Immigration studied the question and issued an interim report in October 2003 (House of Commons Canada 2003b). As part of its study, the Committee reviewed prior considerations for a national identity card, reviewed polling results, held hearings, and visited a number of European countries already using or planning to introduce national identity cards. Although the interim report of the Committee made no specific recommendations for or against the concept of a national identity card, it noted that the majority of witnesses before the Committee came out strongly against the idea of a national identity card system.^{Footnote 7} The Committee also identified a list of questions that would have to be answered before the concept could proceed. As of this writing, there do not appear to be any plans for the government to introduce a population registration program.

Statistics Canada has an Address Register (AR). Coverage of the AR is very good in urban areas, but weaker in rural areas. The AR was designed with mail-out of census questionnaires as its primary objective, and is not a true building/dwelling register. As well, a major source of updates for the AR is the traditional census itself.

Statistics Canada's Business Register (BR) is generally of high quality, but is primarily designed for conducting business surveys. It excludes some small unincorporated non-employer businesses, although such businesses are not economically significant in most industries. The BR does not appear to have been evaluated as a source of data (e.g., place of work) for individuals.

---

Other countries:

All countries using the register-based approach first developed population registers for administrative, not statistical, purposes (ONS 2005b, Redfern 1989). In the cases of buildings and dwellings, additional data collection has sometimes been needed to construct a dwelling register (e.g., Norway and Sweden), and a method of keeping it up to date is required.

The experience of other countries is that once a central population register is established, it can be used in the census within 5 to 10 years, but full elimination of data collection can take several decades. In some cases, some census content may need to be dropped (e.g., occupation and languages spoken, in the case of Austria). In other cases (e.g., Germany, Switzerland and the Netherlands), the register-based approach must be supplemented by additional data collection, which may itself be mandatory.

The creation of a National Population Register in India is undoubtedly the most ambitious project of this type ever attempted. At this point how successful it will be is unknown and its role in the 2021 Census, if any, is yet to be defined.

Canada:

Canada has no population register and no universal legal requirement for Canadian residents to register changes of address with the government. Other methods for detecting address changes (e.g., voluntary reporting or using administrative files, such as tax records or driver's licence files, as a source of address updates) might be considered, but further investigation would be needed to determine whether the quality of address updates would be acceptable for statistical purposes.

Addresses on the AR and businesses on the BR are generally geocoded to the small-area level. For the AR, the geocoding is based on the data holdings in the geographic databases and information from the previous census. In the case of the BR, an assessment of the quality of the geocoding would likely be required to determine whether it would be sufficiently accurate for census purposes.

---

Other countries:

In countries with population registers, it is generally the case that the population is required by law to register and to report any change of address to the register authorities. Geocoding is normally accomplished through the building/dwelling register and in some cases through the business register. Persons are linked to the dwelling register through their PIN, thus allowing the geocoding of persons and the determination of the household composition. The central population register may also contain references to parents and spouses, allowing families to be identified and geocoded. Linkages to the business register allow the place of work to be geocoded.

Canada:

Records of births and deaths of persons are generally reliable and are provided to Statistics Canada by the provinces and territories, with the timeliness dependant on their administrative processes. Immigration records are comprehensive, although some of the associated information (e.g., address in Canada, date of entry) may be imprecise. There is no legal requirement for persons leaving the country to report their departure, nor is there any requirement to report internal migration.

The AR is currently updated in advance of the census by field-canvassing methods, and again after the census using the census results. Administrative sources, such as telephone files, are used to update the AR between censuses.

The BR is maintained using administrative data from the Canada Revenue Agency and survey feedback. There are some delays in recording events; particularly problematic is the identification of businesses that have ceased to exist.

---

Other countries:

'Births' and 'deaths' of units (persons, addresses and businesses) are among the main events of interest. The Nordic countries report that in practice it is often difficult to pinpoint the real date of an event , as distinct from the date on which the event was recorded (UNECE 2007, paragraph 125).

Prompt reporting of events to the register authorities is the norm due to its mandatory nature. For example, in the Netherlands all births must be registered within four working days, otherwise the parents receive a fine. In Finland the allowable period to report a change of address is seven days.

Canada:

To produce small-area census data in the same time frame as the 2016 Census or NHS, a continuous sample survey would have to start full-scale data collection in early 2012, in order to accumulate sufficient sample by 2016. The sample size of such a survey would depend on whether it would be used to replace only the long-form data (as in the United States) or as a complete replacement for the 2016 Census (as in France).^{Footnote 8}

At this point, no development or testing of a continuous survey to replace all or part of the census has been conducted, nor does Statistics Canada appear to have made any assessment of the amount of time required to do so. Based on my experience, it is very unlikely that such a survey could start collecting data any earlier than mid-decade, even if development work were to start almost immediately. There is no possibility that data collection could begin by 2012.

---

Other countries:

The United States took 18 years (from 1993 to 2010 inclusive) from the research proposal stage to the first publication of five-year estimates. The lessons learned from the pioneering efforts of the U.S. Census Bureau could shorten this period for Canada to some extent, but Canada would likely require the same four phases of development, testing, demonstration and full implementation.

In particular, the United States implemented the ACS only after a comparison of the 2000 Census long-form data to the corresponding data from the demonstration ACS, dropping the long-form content from the Decennial Census only in 2010. Because much of the value of census data is in tracking long-term trends, such a 'parallel run' in the 2016 Census might be considered indispensible by data users in order to understand the effects of such a major change in methodology.

In a scenario where a demonstration phase would be conducted in parallel with the 2016 Census, the development and testing phases would have to be completed by 2015. For the continuous survey to generate small-area long-form content in the time frame of the 2021 Census, the survey would have to start full-scale data collection by 2017. Statistics Canada would have to assess in further detail its ability to meet such a schedule while also conducting a full 2016 Census and/or NHS, but at this point, such a scenario would appear to be very high risk.

The U.S. Census Bureau also conducted a test of voluntary response to the ACS in 2003. The test showed the results of making response to such a continuous measurement survey voluntary: lower response rates, lower data accuracy, and higher costs to maintain the same level of quality.

The development of the rolling census in France took 12 years (1997 to 2008 inclusive) from the initial proposal stage to the first full production of results. The approach is still new, and is in a phase of evolution as lessons are learned and incorporated into the methodology. Nevertheless, it appears to have achieved its primary goal of making data available on a more regular and timely basis.

Canada:

There does not appear to have been any broad stakeholder consultation in Canada on a continuous measurement approach to collecting census data. Canada has conducted a national census every five years since 1956,^{Footnote 9} so the trade-offs for users in Canada may be quite different than for countries such as the United States and France, where the traditional censuses previously occurred every 10 years (United States) or at irregular periods of 6 to 9 years (France).

---

Other countries:

The main driver behind the ACS was the need to improve the timeliness of small-area socioeconomic data compared to the Decennial Census. The U.S. Census Bureau has expended extensive efforts on consultation and training of data users on the new type of data.

For example, users need to be taught that although multi-year estimates from a continuous measurement approach are updated annually, this really means that the oldest year of data is dropped and replaced by the newest year, i.e., the estimates are not based on completely new data.^{Footnote 10} In addition, it is important that users choose the correct set of estimates to use when analyzing data. Large geographic areas may have one-year, three-year and five-year estimates available while small areas only have five-year estimates; in such a case, the user would have to know to use the five-year estimates for all areas if the estimates are to be compared.

In France, legislative changes were required to move to an approach based on continuous measurement over a five-year rolling period and on the production of official population figures using a 40% sample, instead of a complete enumeration, in the larger communes. The French approach avoids the problems of having estimates for time periods with different lengths (all estimates are based on five years of data), but this requires a considerable degree of mathematical modelling. Canada would have to assess the extent to which such modelling would be legally acceptable in the Canadian context, and possibly modify the approach.^{Footnote 11} The French approach also means that even the estimates for large geographic areas are based on five years of data, even though the sample sizes for such areas might be sufficient to produce reliable one-year estimates.^{Footnote 12}

Both the United States and France have found that a substantial amount of user support is required to interpret the data. In both cases, the approaches are fairly new, and lessons are still being learned. Any user consultation would have to carefully explain the strengths and weaknesses of updated time-period averages or estimates that involve a substantial element of statistical modelling.

Canada:

Continuous updating of the AR is a long-term goal, and significant progress has been made in the past few years to develop and use administrative sources and targeted field canvassing operations, with the eventual goal of supporting the ongoing household survey program as well as the census. However, a major source of updates to the AR remains the results of the census itself. Considerable development and testing would appear to be needed to determine whether updates could be made from other sources alone that would support a large-scale ongoing survey program.

Alternatively, area sampling methods similar to those used in the current Labour Force Survey could be used. However, this would result in a less efficient sample and therefore a higher required sample size and increased respondent burden, due to the multi-stage methods inherent in an area sample approach. Area samples also deteriorate over time and need to be refreshed occasionally, based on an up-to-date census.

---

Other countries:

Both the United States and France made major investments to upgrade and maintain their address lists. In the case of the United States, their Master Address File covers (in theory) 100% of the country. In the case of France, the lists are not yet of sufficient quality in the smaller communes, so a rotating census of communes is used instead of sampling addresses directly, as in the larger communes.

The French approach is very dependent on the administration and infrastructure context that is specific to France. Using such an approach in Canada would likely require adaptation, in particular to take account of the differences in the geographical dispersion of the population and in the administrative responsibilities for data collection in Canada. Considerable development and testing would probably be required.

Canada:

As noted above, using a continuous survey to produce small-area data in a time frame equivalent to a 2016 Census long form or 2016 NHS would require that the continuous survey begin collecting data by 2012. Such a start date is not possible, and at this point, Statistics Canada does not appear to have formally assessed what a feasible start date could be.

---

Other countries:

In the case of the ACS, it took one, three and five years of data to produce estimates for, respectively, areas of 65,000 or more, areas of 20,000 or more, and all other areas. In the case of France, it took five years of data collection before the first estimates were produced, for all areas.^{Footnote 13} However, all estimates can now be updated annually.

Canada:

The current census funding model is based on supplementary funding, with a peak in census year. Funding is under the control of the federal government, and could presumably be allocated in a different manner, but at the present time no funding of any kind appears to have been allocated for the development of a continuous measurement approach.

---

Other countries:

In the case of the ACS, there was a two-year delay in implementing the approach due to budget problems, although funding now appears to be stable. Because the Canadian census is required to be conducted in specific years by the Constitution Act and the Statistics Act, Canada has not experienced delays in conducting the census for budgetary reasons. If Canada moved to a continuous survey that was legally separate from the census, the continuous survey could become subject to such budgetary uncertainties.

In the case of France, the postwar traditional censuses were often postponed due to budgetary problems. The new rolling census approach, together with the new legislation underlying it, spreads the costs more evenly and has resulted in more consistent data production.

Canada:

In principle, Statistics Canada has the professional capacity to undertake the development of a large-scale continuous sample survey. However, an important part of this capacity is allocated to the 2011 Census and the 2011 NHS. Under current conditions, it appears unlikely that these employees could begin development work on a continuous sample survey approach before 2012 or 2013, when work on the 2011 Census and the 2011 NHS respectively is completed.

---

Other countries:

The U.S. Census Bureau and INSEE have had the professional capacity to develop and maintain a continuous measurement approach. The key issue is how this capacity was organized in order to switch over to a new approach.

In the case of the United States, two separate organizations within the U.S. Census Bureau were established to develop the ACS and the 2000 Decennial Census. Once the ACS was fully implemented, the ACS and the Decennial Census became part of the same directorate.

In the case of France, a separate organization was set up within INSEE in 1997 to develop the rolling census, at the same time that the 1999 Census was being carried out. From 2001 to 2004, the organization grew into a special project team reporting directly to the director general of INSEE, given the challenges and implications for INSEE of the new approach.

If Statistics Canada were to develop an approach involving continuous measurement at the same time as it was planning and conducting the 2016 Census and/or the NHS, it would likely wish to establish a separate organization to do so, as the U.S. Census Bureau and INSEE did.

The ability of users to interpret time-period average data instead of point-in-time data is still in evolution. The United States published its first five-year estimates only recently (December 2010).