Potential approaches for the 2021 Census and beyond

The primary purpose of this study was to identify and provide an initial assessment of methodology options for the 2016 Census. The study also helped to identify potential approaches which, although not feasible for 2016, could be considered for the 2021 Census or beyond. In this section, I provide a brief overview of three potential approaches for this longer time frame. Considerably more work would be needed by Statistics Canada to develop and assess these approaches in detail before decisions about the census methodology for 2021 and beyond could be taken.

A register-based census based on a new Central Population Register

For Canada to conduct a future register-based census modelled on those conducted in countries with administrative registers, the following developments would likely be needed.

First, a government-wide business case for establishing a Central Population Register (CPR) would be required. The question of how much such a CPR would cost to establish and maintain cannot be fully answered at this point, but some estimates of the magnitude of the costs are available from existing sources. A 1999 study (HRDC 1999) estimated the cost of establishing a 'Common Client Identifier,' supported by biometrics technology, at between $1.1 billion and $3.6 billion. The 2003 submission of the Interim Privacy Commissioner to the Standing Committee on Citizenship and Immigration estimated the start-up costs of a national identity card system at $3 billion to $5 billion.^{Footnote 1}

The cost of the 2011 Census/NHS program is approximately $660 million, so it is unlikely that a CPR would be cost-effective solely on the basis of replacing the census. A CPR would need to provide additional offsetting savings and/or benefits to the government, such as security enhancement, fraud prevention and administrative efficiencies. Statistics Canada would probably not be capable of assessing these aspects of the business case.

As part of the establishment of a CPR, the government would need to adopt a universal (and therefore mandatory) PIN as well as legislation permitting the use of the PIN for linking the CPR and other administrative registers for statistical purposes. Such legislation should also include data protection measures. A universal PIN might be established by making the existing SIN universal, or, as in Switzerland and India, by issuing a new PIN to all Canadian residents (both permanent and non-permanent).

The reporting of changes of address to the register authorities would ideally be mandatory (e.g., as currently required for a driver's licence) in order to keep the CPR up to date, so that it would be useful for both administrative and statistical purposes.

The acceptability to both the public and census stakeholders of using the CPR to conduct all or part of the census would also have to be established. The privacy impacts and public acceptability of trading off mandatory registration and reporting of changes of address with no longer having to complete part of the census would need to be determined.

If a CPR were to be established, it could likely be used as the basis of the census within 5 to 10 years. To eventually move to a register-based only census, additional registers (e.g., education, dwelling, employment, families) would have to be developed to cover content that is not available in existing registers. The feasibility and the costs to develop these registers would need to be assessed. The experience of other countries suggests that it could be several decades before the content and the data quality of such registers would be sufficient to adopt a register-based only census; in the meantime, the CPR would have to be supplemented by surveys and/or complete enumerations to collect the additional content.

Finally, some of the content that is currently collected in the census or the NHS might never be part of any register, for example ethnic origin, visible minority status or mode of commuting to work. Surveys would still be required if data on such topics are important. The costs to conduct such surveys would have to be assessed.

The major cost and privacy implications of these developments make it appear unlikely that they would occur in time for the 2021 Census. Because of this, I recommend that the approach not be pursued for 2021.

Keeping the census up to date through administrative data and surveys

A second potential approach is to keep the results of a traditional census up to date by using a combination of existing administrative files and surveys, without the construction of a Central Population Register and the creation of a PIN.

One scenario for the 2021 Census might be to begin with the 2016 Census as the base population, and attempt to update it with birth, death, immigration and other records that capture demographic and other changes. Changes of address of individuals might be updated from existing administrative records (e.g., tax files) and linked to dwellings via the Address Register. Statistics Canada might also exploit the expertise it has developed in forming families in the T1 Family File^{Footnote 2} to create families and households.

A large sample (possibly several million households) might then be selected from this updated population and household database and intensive attempts made to trace and interview the individuals, using the expertise developed in the census coverage measurement studies and longitudinal surveys. The results of the survey might then be used to adjust the database using small-area estimation techniques,^{Footnote 3} for example, for over- or under-reporting of address changes, emigration or changes in household composition. The sample survey might also be used to collect updated socioeconomic data and to adjust the database, again perhaps using small-area estimation techniques.

To the best of my knowledge there are no countries that actually use such an approach for conducting a census. Similar projects have been attempted on an experimental basis or considered in a few countries that do not have population registers. The U.S. Census Bureau conducted an Administrative Records Experiment (AREX) as part of their 2000 Census research program and is expanding the scale and scope of this research for its 2010 Census. France has developed a longitudinal database (Échantillon Démographique Permanent) of persons born on certain days of the year, and links demographic events (births and deaths) to the census forms. The United Kingdom has been considering an administrative records approach as one option for their 2021 Census.

If such a project were to be attempted, Statistics Canada should initiate research and testing relatively soon so that the methodology could be developed, administrative sources could be identified and linked, and samples could be selected and traced. In order to determine the feasibility of such a method for the 2021 Census, one could start with the 2011 Census, develop and apply the methodology, and compare the results of this approach to those of the 2016 Census. Even if the conclusion were to be that such an approach to conducting a census is not feasible in time for the 2021 Census, the lessons learned in the attempt could be very worthwhile in the longer term for increasing the use of administrative data in both the census and the household survey programs.

As part of such a project, it is suggested that Statistics Canada and the government should also confirm that no legislative changes are needed to permit existing identification numbers such as the SIN to be used to link such files, thus improving the quality of the linkage, and that additional administrative files, such as provincial or territorial health care files, could be used in such an approach. It is also recommended that the public acceptability of linking the census, administrative records and tracing surveys together to produce census-type data be assessed through public consultations prior to implementing such a methodology.

Continuous measurement approaches

Although a continuous measurement approach as used by the United States or France is not feasible in the time frame of the 2016 Census/NHS, it might be a feasible alternative for 2021 or beyond. A continuous measurement approach could become simply the source of long-form data, with a traditional short-form census still conducted every 5 or 10 years or, at the extreme, could potentially replace the census completely. It might be noted that a continuous measurement approach has no inherent advantage over a point-in-time census in terms of response burden or intrusiveness; in both cases, data are collected from the public. The primary motivation for a continuous measurement approach is improved timeliness.

As a first step, Statistics Canada and the government would likely need to assess the legal implications of the continuous measurement approach , particularly with respect to the legal requirement to conduct a census in specific years. For example, it would need to be determined whether continuous measurement could be considered to be part of the census under the relevant legislation. This determination would have implications for which questions could be part of the continuous survey or the traditional census and whether the continuous survey could be voluntary or not. The legal implications for the extreme form of continuous measurement, which departs completely from the current approach by replacing a census taken in specific years with continuous data collection, might be particularly significant.

Assuming that the legal implications would not be a barrier, the next step would be for Statistics Canada to consult users and other stakeholders to determine whether a continuous measurement approach would have any significant net advantages compared to the current approach of a census conducted every five years. The advantages and disadvantages (e.g., timeliness, frequency, averages versus single point-in-time estimates) would have to be carefully explained, and there should be widespread user agreement before proceeding. The role of a continuous sample survey within the broader context of the overall household survey program would also need to be assessed; for example, the uses of such a survey as a sampling frame for other surveys.

In parallel with the stakeholder consultations, Statistics Canada would need to conduct a feasibility study, including costs and timelines, for developing, testing, demonstrating and implementing a continuous measurement approach. In particular, Statistics Canada would need to estimate and compare the ongoing costs in a Canadian context of the traditional census approach, the traditional census with yearly updates of characteristics approach and the rolling census approach. Only one country has implemented each of the latter two approaches, and it would not be reasonable to simply extrapolate their cost structures to the Canadian context.

Finally, the new methodology would need to be developed, tested, demonstrated and implemented. Although Statistics Canada could benefit from the experiences of the United States and France, it would likely take at least 10 years for such a program and would require a significant long-term commitment of resources.