The Demosim microsimulation model at Statistics Canada: A tool for policy planning and evaluation

Abstract

This paper presents an overview of Demosim, a demographic microsimulation projection model developed at Statistics Canada to support policy planning and evaluation. We first outline the circumstances that initially led to the creation of the model, the challenges met during its development, and the advantages and complexities of using such a model. We then demonstrate, through several concrete examples of studies that used Demosim, the power and versatility of the model to answer a variety of research questions, thanks to its capacity to project a large number of characteristics and complex demographic processes. We show how Demosim results have contributed, for instance, to providing the public and policy makers with detailed projections that reflect the future ethnocultural diversity of the population or focus on the Indigenous populations in Canada. Finally, we provide an in-depth illustration to analyze how theoretical changes in the education of current school-age cohorts would impact the size and qualifications of the future workforce among Indigenous populations. Results show that through the process of demographic metabolism, the workforce would be renewed by more educated cohorts in a slow and gradual process as successive young school-age cohorts yet have to graduate and enter the labour market.

Acknowledgments

The authors want to thank Laurent Martel, who reviewed a preliminary version of this document and provided comments. They are also grateful to David Pelletier, who helped them establish the structure of the article and provided helpful advice.

Introduction

Most statistical agencies use cohort-component models to produce population projections by age and sex and a multistate cohort component to incorporate additional dimensions, most often regions (multiregional model). Multistate models use a matrix framework based on simple yet clever mathematics, but this framework becomes quickly impractical when the number of additional dimensions is large. This goes against the ever-increasing appetite of policy makers, governmental and non-governmental institutions, and the public for actionable population projection data with high granularity, like what is found in census data, for example. Of particular importance from a policy planning and evaluation perspective are data on the ethnocultural diversity present in many countries, whether stemming from long-standing population heterogeneity, colonization, or more recent international migration patterns. Microsimulation offers a much more flexible alternative for highly disaggregated population projections. By making individuals, rather than age, sex and region aggregates, the unit of analysis, microsimulation can project a much larger number of characteristics than traditional multistate models. The development and use of microsimulation models, however, entail large investments that become interesting only if models are long-lasting, adaptable, and relevant for decision making and if data exist to supply the model with the numerous parameters that it needs.

In this paper, we present an overview of Demosim, an ever-evolving demographic projection microsimulation model developed and maintained at Statistics Canada for more than two decades. We first outline the circumstances that initially led to Demosim’s creation and the challenges met during its development. We then present several concrete examples of studies done using Demosim that are based on either plausible or “what-if” scenarios, long-term projections, or nowcasting approaches. We show how results from Demosim have contributed to informing public debates with detailed projections that reflect the future ethnocultural diversity of the population or focus on the Indigenous populations in Canada. We follow with a detailed illustration where Demosim was used to analyze how theoretical changes in the education of current school-age cohorts would impact the size and qualifications of the future workforce among Indigenous populations. Results show that through the process of demographic metabolism, the workforce would be renewed by more educated cohorts in a slow and gradual process as successive young school-age cohorts yet have to graduate and enter the labour market.

In reviewing these findings, the paper demonstrates how such microsimulation population projections can provide policy-relevant insights on the mechanics of Canadian demographic dynamics and on major demographic theories, such as demographic metabolism, transition and dividend.

Genesis of Demosim, a population microsimulation model at Statistics Canada

In 2004, Canadian Heritage (PCH)—a Canadian federal department whose mandate aims at fostering and promoting “Canadian identity and values, cultural development and heritage” (PCH, 2020)—initiated a project to draw a portrait of the Canadian ethnocultural situation in 2017 in preparation for the celebrations of the 150th anniversary of Canadian Confederation. One facet of this project was to produce a demographic profile of the Canadian population in 2017 based on variables such as visible minority status, immigrant status, religious denomination, and mother tongue (Bélanger & Caron Malenfant, 2005). The realization of this mandate led Statistics Canada to develop a microsimulation model named Popsim in its first iteration and now known as Demosim. Free from the matrix infrastructure inherent to the cohort-component model typically used by national statistical agencies and by Statistics Canada since the mid-1970s, this microsimulation model provided the flexibility to produce disaggregated population counts for many individual characteristics and geographies.

Statistics Canada’s long experience in the conception of microsimulation models, beginning with the development of the Social Policy Simulation Database and Model (Wolfson et al., 1989) and Demogen models in the 1980s (Wolfson, 1987), but also the POpulation HEalth Model (POHEM) in the early 1990s (Wolfson, 1994) contributed to making microsimulation a viable option. Starting from scratch would have been a much more perilous endeavor, in part because significant advances and breakthroughs were made within Statistics Canada during these two decades of microsimulation models development prior the birth of Demosim. One major milestone was the development in 1993 of Modgen^Note , a generic software environment specifically designed to facilitate microsimulation model programming. Modgen removed different types of obstacles to microsimulation model creation, maintenance, and development. Modgen provided a graphical user interface as well as a simulation engine, allowing new processes to be added without changing the model. The arrival of Modgen has enabled the development of models with a more flexible design, thanks to their modular nature. In addition, Modgen contained tabular language that reduced the memory required during simulation by enabling on-the-fly tabulations. In short, Modgen helped the maintenance and the development of the different microsimulation models at Statistics Canada. At the same time, Modgen has contributed to the expansion of the microsimulation community within Statistics Canada and externally as well, by enabling analysts and scholars to become directly involved in the development of microsimulation models, without having to rely on programmers to program their models. The microsimulation tool has thus spread to the agency’s various fields of analysis, including demography.

Nevertheless, the use of the microsimulation tool is subject to certain conditions and prerequisites. And more globally, the development and operation of a microsimulation model require a large spectrum of resources. This first requires a number of qualified individuals with different skills, including microsimulation experts, developers and subject-matter experts, often specialized in a variety of fields. Powerful computational resources are another prerequisite to running complex simulations. All these requirements make development and maintenance costly. Luckily, over the years, Demosim could count on the financial contributions of several federal departments and other organizations for its development and operation. Finally, the creation of a detailed base population and the consistent modelling of numerous events make microsimulation models data hungry. Demosim benefits from the availability of numerous high-quality demographic data sources in Canada, such as censuses, surveys, and vital statistics databases, enhanced nowadays by data linkages that can often replace longitudinal databases for the measurement of transitions at the individual level.

What is microsimulation and why is it adapted to population projection?

Microsimulation models are simulation models that operate at the level of the individual behavioral entity, in our case people, but it could be a family or a firm. A database comprises a large representative sample of these low-level entities to draw conclusions that apply to higher levels of aggregation (Statistics Canada, 2022a).

For population projections, microsimulation is a sound approach when the heterogeneity of the projected individuals matters. Free of the matrix infrastructure typical to models that project cohorts, microsimulation models can simultaneously, coherently and efficiently project many characteristics of the population (van Imhoff & Post, 1998; Willekens, 2011; Bélanger & Sabourin, 2017). The heterogeneity of individuals may also matter when assuming the same behavior among members of a cohort can compromise the plausibility of the projections because of changes in the composition of cohorts over time (Lutz, 2021).

Microsimulation models are of different types. Demosim was developed as a case-based model, meaning that it proceeds by simulating the life of each entity, in our case individuals, one at a time. This contrasts with a time-based model, where all individuals are projected simultaneously. Each type has its advantages. In a time-based model, it is possible to model interactions between individuals and simulate the formation and dissolution of groupings, such as families or households. Case-based models, on the other hand, tend to require much less computer resources, and this is a crucial advantage for Demosim given the large number of events modelled and the large size of the base population.

Demosim is also a continuous-time model, in contrast to a discrete-time model. In a continuous-time model, the various events modelled in the projection can occur at any moment during the simulation. The occurrence of events is contingent on waiting times based on an individual’s set of characteristics (competing risks). Waiting times associated with events are recomputed each time there is a change in one of these characteristics. In most cases, waiting times are derived from results of statistical models, such as multivariate regression models or contingency tables. Moreover, as Willekens (2009) puts it, “in continuous-time microsimulation, the theory of competing risks determines the timing and sequence. That allows a more accurate study of temporal sequence of events than in discrete time analyses. In addition, the theory allows to model complex event sequences and interactions between events. The fact that microsimulation models in discrete time are not able to handle complex and interdependent event sequences is viewed as an important limitation (Zaidi and Rake, 2001: 19).” Indeed, the mechanics behind discrete-time models are usually simpler, and in these types of models, only one event can occur at a time (e.g., annually). This requires the sequence of events to be predetermined, and the time between events can only be determined approximately. These drawbacks are not negligible when the number of potential events is large.

Demosim is currently programmed with Modgen, a generic microsimulation programming language mentioned earlier which manages nicely mechanisms such as event queuing and the calculation of waiting times. This language, based on C++, supports the creation, maintenance, and documentation of dynamic microsimulation models.^Note  For the future, Demosim will be programmed in OpenM++, an open-source programming language that is cross-compatible with Modgen.^Note 

As demonstrated later in this paper in the context of demographic projections, microsimulation models are particularly versatile and can answer a variety of research questions on different aspects because they handle the projection of many characteristics simultaneously, efficiently, and consistently. These research questions can be easily translated into projection scenarios that are then incorporated, as parameters, into a model.

Demosim: A powerful tool to support policy planning

Demosim is a population projection model built to support decision making and policy elaboration in various Canadian federal departments—such as PCH; Indigenous Services Canada (ISC); Immigration, Refugees and Citizenship Canada; and Employment and Social Development Canada. These have been the main funding partners of the model over time. The projections prepared for these departments are also made available to all other users, including decision and policy makers at the Canadian provincial, territorial and municipal levels; researchers; academics; university students; and the public. Usages have been diverse: projection of the size of the clientele for various types of service delivery, urban planning, conception of employment equity programs, teaching, and research. Over the years, the model was modified to respond to an ever-evolving spectrum of data needs.

Outputs from Demosim are prospective counts like those produced by cohort-component models of population projections but with a much higher level of disaggregation. The projected variables were chosen because they were requested by stakeholders or because they are predictors for the projections of other variables. An example of the latter case is marital status: while not being explicitly requested by Demosim’s clients and end users, this dimension is modelled in Demosim because it is a good predictor of fertility behaviour, improves the quality of modelling of the transmission of some characteristics from mothers to children, and of the attribution of head-of-family and head-of-household statuses. The rich outcomes from Demosim are also used to produce derived projections, i.e., projections made externally to the model, usually at a relatively low cost. Table 1 shows a list of the main projected variables in the most recent versions of Demosim.

Each change in state related to a variable consists of an event and, as such, requires its own component (modelling) in Demosim (e.g., obtaining a degree, changing religious denomination during life course, learning one of Canada’s official languages).

The base population must contain the variables of interest and is derived from the Canadian Census of Population (long-form questionnaire) microdata file. Some adjustments are made to the microdata file to ensure that the base population matches estimates from Statistics Canada’s Demographic Estimates Program.^Note 

Theoretically, the potential of microsimulation’s “expanding content” is limited only by computer resources and the capacity to model the various events and their interactions. However, the complexity of the model increases almost exponentially with each variable added. As the model grows larger, the time to rebase it when new data sources become available greatly increases. This has consequences on the cost and the timeliness of the dissemination of the results and, hence, their relevance. More generally, large and complex models are difficult to maintain over time. Finally, the numerous interactions between variables and events contribute to making validation a tedious and lengthy endeavor. A balance regarding the complexity of the model must therefore be preserved. Demosim has nevertheless been continuously evolving over the years to meet an ever-widening spectrum of data needs, some examples of which are presented in the following section.

The various uses of Demosim

Policy makers are accustomed to thinking in terms of scenarios to evaluate the outcomes of different policies or to guide policy development. Demosim provides the required flexibility to build tailor-made projection scenarios to address many research questions. Broadly speaking, these scenarios tend to serve three purposes, as shown in Table 2.

Scenarios that aim to project a plausible future

Most users, including the main stakeholders funding Demosim, are looking for plausible projections of the Canadian population for policy planning, program management or program evaluation. Key features of the projections are a relevant level of disaggregation and plausibility of the outcomes. The results come with the usual warning that they are not predictions or forecasts, and users are informed of the limits in how a projection model, even as complex as Demosim, can foresee the future. To reinforce this statement, Statistics Canada always develops several scenarios (Dion & Galbraith, 2015). A plurality of scenarios highlights the uncertainty inherent to projections and shows that there are several plausible paths the future could take (United Nations Economic Commission for Europe, 2018).

Four main series of projections have been developed over the years for users looking for projected population counts from plausible scenarios about the future: (1) projections aiming to reflect the future diversity of the Canadian population, (2) projections of the Indigenous populations and households in Canada, (3) language projections for Canada, and (4) labour force projections for Canada.

The future diversity of the Canadian population

The first projections aiming to reflect the future diversity of the Canadian population were published in 2005 (Bélanger & Caron Malenfant, 2005). They were based on the 2001 Census and covered all provinces and territories. Since then, three updates to these projections were produced following each subsequent census, with the addition of new variables and a more granular geography covering all largest urban centers of the country (Caron Malenfant et al., 2010; Morency et al., 2017; Statistics Canada, 2022b). The various editions of these projections dedicated to the future diversity of the Canadian population have provided key information to help capture how the social fabric of Canadian communities is expected to change in the future.

For example, the most recent edition contains 11 distinct scenarios, each proposing plausible variations regarding one dimension influencing the future evolution of the ethnocultural diversity in Canada, such as the annual intake of immigrants, place of birth of immigrants, geographical distribution of new immigrants at landing, fertility, mortality, internal migration, and number of non-permanent residents. Projected counts are available by immigrant status, generation status, visible minority group, age, sex, and place of residence. Users may compare the various scenarios to get a better understanding of the factors impacting the ethnocultural diversity of the Canadian population and how this evolution may differ regionally. The results showed that half of the Canadian population could be made up of immigrants and their Canadian-born children in 2041, up from 38.2% in 2011. The proportion of the population belonging to a visible minority group could reach 40%, almost twice the proportion observed in 2016. However, because immigrants tend to settle mostly in the largest urban centers, the contribution of immigrants to population growth and the relative increase in the number of Canadians with diverse backgrounds and cultural influences are expected to be very unequal across the country.

David Coleman (2006) talked of a third demographic transition to depict the societal transformation occurring in countries with sub-replacement fertility and high levels of immigration, particularly migration from less developed countries. These transformations have important implications in terms of historical identities, social cohesion, and policy planning. Demosim has been used to describe some aspects of these transformations, thanks to its capacity to mimic the two distinct processes behind the ethnocultural diversification of the population: the continuous arrival of new immigrants and the transmission of characteristics from parents born outside Canada or with some foreign-origin ancestry to their Canadian-born children. For example, Dion et al. (2014) used special simulations in Demosim to find out how fast a third demographic transition is occurring in Canada by looking at the share of immigrants and the descendants of immigrants under various projection scenarios that differ from one another according to their assumed levels of immigration and mixed unions. Assuming the entire population was born in Canada in 2006, the study showed that under the existing regimes of immigration, it would take between 41 and 55 years for the proportion of new immigrants and descendants of new immigrants to reach 37%, the proportion observed in 2006. This is a relatively short time for a society to adapt to sizable changes, stretching over only two generations.

Projections of the Indigenous populations and households in Canada

The projections of the Indigenous populations and households in Canada were developed following a request by ISC, whose mandate is to improve access to high-quality services for First Nations, Métis, and Inuit populations. Projected counts are useful to estimate the future population sizes and enable rigorous planning of services and their costs. The capacity of Demosim to model many variables and events is useful because the growth of the First Nations, Métis and Inuit populations is driven not only by purely demographic factors but also by response mobility^Note  (O’Donnell & LaPointe, 2019) and new registrations to the “Indian Register” because of legislative changes to the Indian Act (Crown-Indigenous Relations and Northern Affairs Canada, 2018).

The first edition was published in 2011 (Caron Malenfant & Morency, 2011) and was followed by two subsequent editions (Morency & al. 2015; Statistics Canada, 2021). The most recent results from Demosim show that the Indigenous population in Canada, estimated at 1,800,000 in 2016, could reach 2,495,000 in 2041 under the low-growth scenario, 2,848,000 under the medium-growth scenario and 3,182,000 under the high-growth scenario. The Indigenous population is projected to grow faster than the non-Indigenous population, despite the record-level number of immigrants coming to Canada and contributing to its growth. As a result, the share of the Indigenous population in the overall Canadian population could range from 5.4% to 6.8% in 2041, compared with 5.0% in 2016. Two main factors explain the high growth rates of the Indigenous population in the coming years (of varying importance among First Nations, Métis, and Inuit populations): a high fertility and a sustained response mobility.

Language projections for Canada

The high number of international immigrants admitted to Canada each year and their geographic and linguistic makeup have a direct impact on the demolinguistic composition of the population. They also affect the balance between English and French across the country. A key question is the demographic weight of Canada’s French-speaking population, especially given that it historically receives less than its share of the immigrant population.

In 2017, Statistics Canada released the results of 10 language projection scenarios that aimed to provide a realistic portrait of how the two Canadian official languages, English and French, could evolve in Canada between 2011 and 2036 (Houle & Corbeil, 2017). With Demosim, it was possible to account for multiple factors influencing the demolinguistic portrait of communities, such as characteristics of new immigrants, language transfers by parents to their children and patterns of internal migration. The results have informed the Government of Canada’s Action Plan for Official Languages – 2018-2023: Investing in Our Future, which aims to promote the vitality of official language minority communities and bilingualism in Canada (Government of Canada, 2023). They showed, among other things, that if recent trends were to continue, French-speaking people would continue to lose ground as a share of the population, particularly outside Quebec, where the share of French-speaking people is projected to decline, and that a key challenge facing the French-speaking population outside Quebec lies in its relatively old age structure.

Labour force projections for Canada

In recent years, following decades of growth, the overall Canadian labour force participation rate began to decline from 67.6% in 2008 to 65.6% in 2023 (Statistics Canada, 2024a). During this period, the participation rates of older persons and women in general increased significantly. However, these increases were completely mitigated by the aging process of the Canadian labour force. The future development of the labour force is a key issue for policy planners because a larger share of economically inactive individuals translates into important pressures on welfare programs, such as old-age pensions and health care.

In 2024, Statistics Canada released labour force projections under various scenarios of population growth and participation rates by age (Vézina et al., 2024). These are derived projections, i.e. probabilities of labour force participation have been applied annually to populations projected using Demosim. Labour force participation probabilities are broken down by several individual characteristics, including age, sex, education, place of residence, immigrant status, visible minority group and Indigenous identity, among others. Because labour force status depends largely on the demographic characteristics that are projected in Demosim, these derived projections generate sound cross-sectional snapshots of the Canadian labour force. Projections showed the aging process of the labour force would stabilize in the coming years once the baby boom generation has retired. The overall labour force participation rate should continue to decline until the 2030s and would stabilize at around 65%, at least until 2041. The different projection scenarios showed the size and ethnocultural composition of tomorrow’s Canadian labour force is driven mostly by immigration intakes. In fact, the ethnocultural composition of the workforce is projected to change substantially: according to the reference scenario, the foreign-born share of the labour force would continue to rise steadily from 29.7% in 2021 to 43.8% in 2041. However, immigration has limited impact on the future overall labour force participation rate and on the aging and renewal of the labour force. Joining the workforce at the prime of their lives, immigrants help renew the labour force by transforming its ethnocultural composition more than its age structure.

Scenarios developed for sensitivity analysis (“what-if” scenarios)

Sometimes, researchers or policy makers are not looking for projection figures per se but to develop sensitivity analyses to answer research questions. In these cases, the key element is not so much the plausibility of the projected outcomes but rather the capacity to isolate the effect of some specific component of the projections to contrast the results of distinct simulations.

One benefit of sensitivity analysis for researchers is that the process is intuitive and generally well understood by a non-scientific crowd. Because of the large number of variables that can be projected, Demosim can provide answers to various sets of questions covering a wide spectrum of topics. Four examples of research questions that can be translated into “what-if” scenarios and implemented into Demosim are outlined below.

No immigration

What would be the impact on ethnocultural diversification of the Canadian population in the next decades if immigrants were no longer admitted to Canada? The impact of immigration on the actual and future growth of the Canadian population is huge. Immigration almost entirely fuels Canadian population growth, with its share increasing year after year. In fact, natural growth (births minus deaths) is projected to eventually become negative, leaving immigration the sole driver of population growth (Statistics Canada, 2022c). Simulating a complete closure of the Canadian borders therefore would reduce the growth of the Canadian population, which is currently the fastest-growing population among the G7 countries (Statistics Canada, 2022d), to nil.

This stylized scenario reveals that the ethnocultural diversification process of the Canadian population would still go on, regardless of future levels of immigration. Even without immigration, projection shows that the share of people belonging to a visible minority group within the total Canadian population continues to increase. In 2016, visible minorities accounted for 22% of the Canadian population and would continue to increase over the next 25 years to 27% because of the fertility of immigrants already settled in the country and the transmission of some of their characteristics to their Canadian-born children. This highlights how fast the Canadian population is renewed through immigration; recent projections suggest that the proportion of visible minorities within the total Canadian population would reach about 40% in 2041 (Statistics Canada, 2022b).

The impact of immigration on population aging

Immigration can contribute slightly to slow population aging, but it cannot stop the process completely. In most countries, the number of immigrants required to maintain potential support ratios would be beyond past levels and reasonable expectations (United Nations, 2000). How is that so? Caron Malenfant et al. (2011) used sensitivity analysis to dissect the effect of immigration on the age structure of the Canadian population. They developed a series of “what-if” scenarios in Demosim to isolate the various factors by which immigration affects the age structure of the welcoming population: the specific age structure of immigrants, and the fact that immigrants and non-immigrants differ in their propensity to have children, their propensity to emigrate and their life expectancy. The study found that, in the Canadian context, the specific behaviours of immigrants in terms of fertility, emigration and mortality have little impact on the age structure of the population, in part because they act in opposite directions. The rejuvenating effect of immigration is driven mainly by the children of immigrants born in Canada and by the increase of immigration over time. Despite the relatively young age structure of immigrants at arrival, the sole presence of immigrants in the country contributes to the aging of the population in the long term.

Increasing the proportion of French-speaking immigration

What would be the impact on the demographic share of the French-speaking population in Canada outside Quebec if, between 2021 and 2036, the share of French-speaking immigrants among all new immigrants admitted annually outside Quebec was set to various predetermined targets (4%, 6%, 8%, etc.)?^Note The demographic weight of the French-speaking population in Canada outside Quebec has been steadily decreasing between 1971 and 2021, from 6.1% to 3.5%. Many mechanisms are at play, such as natural and migratory increases, transmission of languages from parents to children, and language transfers over the life course. All these aspects are considered in Demosim.

A scenario assuming a continuation of recent trends (about 3% of French-speaking immigrants) shows that the share of French-speaking people outside Quebec would continue to decline until 2036. Setting this share to 4% would not stop this decline but would slightly slow it down. At 6% of French-speaking immigrants admitted annually, the demographic weight of French-speaking people outside Quebec would stabilize, while at 8% or more, the share of this population would increase continuously to 2036. The results of this analysis debunk the myth that the demographic weight of French-speaking people outside Quebec can be sustained with a share of French-speaking immigrants that corresponds to the targeted demographic weight. In fact, the share of French-speaking immigrants among all new immigrants admitted annually outside Quebec must be well above the desired demographic weight of the French-speaking communities to mitigate its decline, which is fuelled by numerous major vectors. These vectors include negative natural growth (more deaths than births) of the French-speaking population, linguistic assimilation in favor of English, the incomplete transmission of French from parents to children, and the low attractiveness of French outside Quebec for immigrants who do not speak either of Canada’s official languages.

No response mobility

What is the contribution of response mobility to the projected growth of the Indigenous population in Canada? The population of Indigenous people in Canada was estimated at 1.8 million in 2016 and is projected to reach 2.8 million in 2041 (Statistics Canada, 2021). In contrast, an (implausible) scenario assuming no response mobility shows that the projected Indigenous population would instead reach 2.2 million in 2041. This means that more than half (61%) of the projected growth to reach 2.8 million can be attributed to response mobility, the remaining share (39%) being driven by other demographic factors. Response mobility is more prominent among certain Indigenous groups, namely Métis and First Nations people. Among Métis, 81% of the projected growth is because of response mobility, while it is 51% for First Nations people.

“Nowcasting” scenarios

In Canada, the Census of Population is the primary data source for estimates for many population characteristics (e.g., immigrant status, visible minority group, generation status, religion, Indigenous identity, knowledge of official languages). It is therefore no surprise that census data are used widely for policy making, program evaluations, market studies, and the calculation of survey weights, among other uses. Census data, however, are collected only every five years, and there is a demand for more timely estimates during intercensal periods, especially given the rapid changes occurring in the Canadian population under high levels of immigration and the strong growth of the Indigenous population. As described earlier, Demosim offers, in some cases, an interesting solution, acting as an extension of the census for current estimates (or very recent past or very near future estimates), which we refer to as “nowcasting.” The rationale for using nowcasting scenarios is simple: in the few years following the release of census data (essentially during the intercensal period), there can be circumstances where the uncertainty associated with a projection can be considered less important than the discrepancy caused by using dated results. A nowcasting scenario is obtained by slightly modifying a pre-existing scenario with the most recent information on the various components of demographic growth (e.g., births, deaths, number of immigrants). In a final step, the results produced by Demosim are calibrated to match population estimates by age, sex, and place of residence for the requested projected date.

Nowcasting projections using Demosim have been produced for many purposes, to include, serving as denominators to compute various types of rates, such as homicide rates for the Indigenous population (David & Jaffray, 2021) and vaccination rates in the context of the COVID-19 pandemic among Indigenous populations and visible minority groups. The results of nowcasting projections are also used internally at Statistics Canada for survey weighting and the validation and certification of census data.^Note  Finally, a nowcasting approach is also used to produce timely targets to be used by the federal government to ensure that its workforce is representative of the various employment equity groups (women, Indigenous people, people with disabilities and members of visible minorities) of the Canadian labour market.

Demosim as a tool for policy making, an illustration

Indigenous people in Canada have, on average, lower labour force participation rates than the non-Indigenous population, specifically Canadian-born White people. According to the 2016 Census, the overall labour force participation rate of the Indigenous population (aged 25 to 64 years old) was 72% compared with 81% for Canadian-born White people. This gap is significant, particularly because the age structure of the Indigenous population is considerably younger,^Note  and this, in turn, boosts the participation rate of a population. One major determinant of labour force participation is education, and Indigenous people are, on average, less educated than Canadian-born White people (Table 3). Moreover, there are significant differences among the different Indigenous identity groups. For instance, the education gaps with Canadian-born White people are largest for Inuit and Registered First Nations and smaller for Métis and Non-Registered First Nations. So, a question that arises naturally is, how does an increase in the educational attainment of Indigenous people in Canada impact their labour force participation over time?

The answer to this question may look deceptively simple: investing in the education of Indigenous people will contribute to increasing their participation in the labour force and a plethora of other benefits. It should, indeed be simple, but as we will see, the devil is in the details, and it is important to see how this would unfold over time to avoid building unreasonable expectations.

Table 3
Education composition by population group, aged 25 to 64, 2016 Census Table summary
This table displays the results of Education composition by population group, aged 25 to 64, 2016 Census. The information is grouped by Population group (appearing as row headers), Bachelor’s degree or higher, High school diploma only, Postsecondary diploma below bachelor level, Total and No high school diploma, calculated using percent units of measure (appearing as column headers).

Population group	No high school diploma	High school diploma only	Postsecondary diploma below bachelor level	Bachelor’s degree or higher	Total
	percent
Canadian-born White people	14	25	37	24	100
Métis	23	27	38	13	100
Non-Registered First Nations	24	28	36	12	100
Registered First Nations	39	23	28	9	100
Inuit	55	18	22	5	100
Source: Statistics Canada, Centre for Demography.

We carried out this analysis using a version of Demosim based on the 2016 Census and projected the population up to 2066, updating and extending an analysis carried out previously by Spielauer (2014). Three scenarios show how alternative assumptions of education trends would affect the educational composition and labour force participation of Indigenous people in the future. Under the base scenario, the education progression follows the general upward trend, and the relative differences in graduation probabilities between Indigenous people and the reference group (Canadian-born White people) are held constant.^Note  The second and third scenarios simulate a convergence of Indigenous people’s education progression rates towards those of Canadian-born White people. Under the phased 50% convergence scenario, the relative differences in graduation probabilities between Indigenous people born in 2006 or later and the reference group are gradually reduced by half. Under the immediate 100% convergence scenario, Indigenous people born in 2006 are set to have the same education progression rates as Canadian-born White people at the beginning of the projection. This unrealistic scenario illustrates what could theoretically generate the fastest closure of the education gap. The exercise is a typical sensitivity analysis. All assumptions other than education remain the same in all three scenarios, suggesting a continuation of recent observed trends. Labour force participation rate assumptions are no exception as they reflect a continuation of the trends observed from 1995 to 2019, showing an upward trend among workers aged 50 and older.

Results for Canadian-born White people reflect the baseline assumption that, following recent upward trends, labour participation rates of older workers would increase in the coming years (Fig. 1). The participation rate is projected to stay stable at around 90% for the younger workers (aged 25 to 44). On the opposite end of the age spectrum, the labour force participation rate of the older age group (aged 45 to 64) is projected to increase. When comparing the baseline scenario with the two alternative scenarios of educational improvements, we find that the pace of the gap closure is rather slow: It would take 20 years after the start of the simulation to notice the impacts of educational improvements on Indigenous people’s labour force participation. Obviously, under the unrealistic scenario of immediate and complete convergence, more pronounced effects are observed. But even in this scenario, changes take time to leave their mark. This is an important result as the impact of new policies or programs is often evaluated after a short period. In this case, an evaluation carried out after five years, for example, would conclude a new policy or program to be a failure. Reasonable expectations would account for the fact that changes cannot occur faster than cohorts are replaced, and this is a slow and gradual process.

Data table for Figure 1

Data for Graphique 1
Projected labour force participation rates by age group and Indigenous identity group under three education scenarios, Canada, 2016 to 2066 Table summary
This table displays the results of Projected labour force participation rates by age group and Indigenous identity group under three education scenarios, Canada, 2016 to 2066. The information is grouped by Age group / Scenario / Indigenous identity group (appearing as row headers), 2026, 2061, 2056, 2036, 2046, 2066, 2016, 2031, 2051, 2021 and 2041, calculated using percent units of measure (appearing as column headers).

Age group / Scenario / Indigenous identity group	2016	2021	2026	2031	2036	2041	2046	2051	2056	2061	2066
	percent
25 to 44 years
Base scenario
Registered First Nations	68	68	69	70	69	69	69	69	68	67	67
Canadian-born White	89	89	90	90	90	90	89	89	89	89	89
Non-reg. First Nations	80	81	82	83	83	83	83	83	83	82	82
Inuit	73	73	74	73	72	72	72	72	72	71	71
Métis	84	84	85	85	85	85	85	85	85	85	84
Phased 50% convergence scenario
Registered First Nations	68	68	69	70	70	71	72	73	73	73	73
Canadian-born White	89	89	90	90	90	90	89	89	89	89	89
Non-reg. First Nations	80	81	82	83	83	83	84	84	84	84	84
Inuit	73	73	74	73	73	74	76	77	78	77	77
Métis	84	84	85	85	85	85	86	86	86	86	85
Immediate 100% convergence scenario
Registered First Nations	68	68	69	70	72	75	77	80	79	79	79
Canadian-born White	89	89	90	90	90	90	89	89	89	89	89
Non-reg. First Nations	80	81	82	83	84	84	85	86	85	85	85
Inuit	73	73	74	73	76	79	82	84	84	84	83
Métis	84	84	85	85	86	86	86	87	87	86	86
45 to 64 years
Base scenario
Registered First Nations	63	64	66	67	68	69	69	69	69	69	69
Canadian-born White	75	76	78	80	81	81	81	81	81	82	82
Non-reg. First Nations	68	70	72	75	76	77	77	77	77	78	78
Inuit	70	72	73	74	75	75	75	74	74	75	75
Métis	72	72	75	77	78	78	78	78	78	79	79
Phased 50% convergence scenario
Registered First Nations	63	64	66	67	68	69	69	69	69	71	71
Canadian-born White	75	76	78	80	81	81	81	81	81	82	82
Non-reg. First Nations	68	70	72	75	76	77	77	77	77	78	78
Inuit	71	72	73	74	75	75	75	74	75	77	78
Métis	72	72	75	77	78	78	78	78	78	79	79
Immediate 100% convergence scenario
Registered First Nations	63	64	66	67	68	69	69	69	72	74	75
Canadian-born White	75	76	78	80	81	81	81	81	81	82	82
Non-reg. First Nations	68	70	72	75	76	77	77	77	78	79	79
Inuit	70	72	73	74	75	75	75	74	78	81	82
Métis	72	72	75	77	78	78	78	78	79	80	80
25 to 64 years
Base scenario
Registered First Nations	66	66	68	69	69	69	69	69	69	68	68
Canadian-born White	82	82	84	85	85	85	85	85	85	85	85
Non-reg. First Nations	74	76	78	79	80	80	81	80	80	80	80
Inuit	72	73	73	73	73	73	73	73	73	73	73
Métis	78	79	80	81	82	82	82	82	81	82	81
Phased 50% convergence scenario
Registered First Nations	66	66	68	69	69	70	71	71	71	72	72
Canadian-born White	82	82	84	85	85	85	85	85	85	85	85
Non-reg. First Nations	75	76	78	79	80	81	81	81	81	81	81
Inuit	72	72	73	74	74	74	75	76	76	77	77
Métis	78	79	80	81	82	82	82	82	82	82	82
Immediate 100% convergence scenario
Registered First Nations	66	66	68	69	71	72	74	74	75	76	77
Canadian-born White	82	82	84	85	85	85	85	85	85	85	85
Non-reg. First Nations	75	76	78	79	81	81	82	82	82	82	82
Inuit	72	72	73	73	76	78	79	80	81	82	83
Métis	78	79	80	81	82	82	82	82	82	83	83
Source: Statistics Canada, Centre for Demography.

Conclusion

The power and elegance of Demosim as a population projection model are linked to its capacity to mimic changes that occur in society through the passage of time over generations, when subsequent cohorts differ from those that precede them. Ryder (1965) coined the expression “demographic metabolism” to describe this process driven by the birth, life and death of individuals. Lutz (2013) developed the concept further of building a “predictive theory of socioeconomic change” that accounts not only for inter-cohort changes but also for intra-cohort transitions. In Lutz’s framework, individuals are the primary building blocks of populations whose behavior will drive social changes (Lutz, 2021). It is easy to see how microsimulation can shed light on and operationalize the process of demographic metabolism.

The case studied in the previous section exemplifies the concept of demographic metabolism. Where studies conducted at aggregate levels could find positive socioeconomic benefits to be gained by increasing the education levels of Indigenous people, Demosim can further demonstrate how these changes unfold in time by accounting for how they occur through cohort succession. Microsimulation can show how changes occurring among cohorts, through various regimes of progression in educational attainment and cohort renewal (a slow process causing important inertia) contribute to social and population changes. We find that even when hypothesizing a complete and immediate convergence of education progression rates of current school-age Indigenous people towards those of the non-Indigenous population, it takes nearly two decades to see notable changes. The projected short- and long-term trends are embedded in the current age and education structure of the Indigenous population, which will change slowly as new cohorts replace older ones.

It is important to recognize that microsimulation is not an optimal solution to all population projection enquiries. There are cases where population heterogeneity does not matter too much, and it is sufficient to project an aggregate of individuals. Population projections by age and sex, for example, are doing well by simply projecting cohorts by age or age group. Moreover, there are costs to using microsimulation, some of which we discussed briefly in this paper, such as monetary costs, complexity, computer resource requirements and difficulties in finding well-versed analysts. Other disadvantages are the lack of transparency of the model caused by, among other things, the large number of parameters and their interactions and the variance induced by the Monte Carlo process necessary to trigger the occurrence of the various events to the simulated individuals. However, as the examples provided in this paper convincingly demonstrate, the potential of microsimulation as a tool for developing policies, conducting research, and answering complex questions is huge.

References

Bélanger, A. & Caron Malenfant, É. (2005). Population Projections of Visible Minority Groups, Canada, Provinces and Regions, 2001 to 2017. Statistics Canada catalogue number 91-541-X.

Bélanger, A. & Sabourin, P. (2017). Microsimulation and Population Dynamics, an Introduction to Modgen 12. The Springer Series on Demographic Methods and Population Analysis. Dordrecht: Springer. https://doi.org/10.1007/978-3-319-44663-9 .

Canadian Heritage (2020, March 10). Raison d’être, mandate and role - Canadian Heritage. Consulted August 24, 2022. https://www.canada.ca/en/canadian-heritage/corporate/mandate.html.

Caron Malenfant, É., Dion, P., Lebel, A. & Grenier, D. (2011). Immigration et structure par âge de la population du Canada, quelles relations? Cahiers québécois de démographie, 40(2), 239-265. https://doi.org/10.7202/1011541ar.

Caron-Malenfant, É., Lebel, A. & Martel, L. (2010). Projections of the Diversity of the Canadian Population, 2006 to 2031. Statistics Canada catalogue number 91-551-X. https://www150.statcan.gc.ca/n1/pub/91-551-x/91-551-x2010001-eng.htm.

Caron-Malenfant, É. & Morency J.-D. (2011). Population Projections by Aboriginal Identity in Canada, 2006 to 2031. Statistics Canada catalogue number 91-552-X. https://www150.statcan.gc.ca/n1/pub/91-552-x/91-552-x2011001-eng.htm.

Coleman, D. A. (2006). Immigration and ethnic change in low-fertility country: A third demographic transition. Population and Development Review, 32(3), 401-446. https://doi.org/10.1111/j.1728-4457.2006.00131.x

Crown-Indigenous Relations and Northern Affairs Canada (2018, November 28). Background on Indian registration. Consulted December 20, 2022. https://www.rcaanc-cirnac.gc.ca/eng/1540405608208/1568898474141.

David, J.-D. & Jaffray, B. (2022). Homicide in Canada, 2021. Juristat. Statistics Canada catalogue number 85-002-X. https://www150.statcan.gc.ca/n1/pub/85-002-x/2022001/article/00015-eng.htm .

Dion, P. & Galbraith, N. (2015). Back to the future: A review of forty years of population projections at Statistics Canada. Canadian Studies in Population, 42(1-2), 102-116. https://doi.org/10.25336/P69K6X .

Dion, P., Caron Malenfant, É., Grondin, C. & Grenier, D. (2014). Long-Term Contribution of Immigration to Population Renewal in Canada: A Simulation. Population and Development Review, 41(1), 109-126. https://doi.org/10.1111/j.1728-4457.2015.00028.x.

Fédération des communautés francophones et acadienne du Canada (2022, March). Faire le point sur la cible en immigration francophone – Bilan, enjeux et pistes d’action. Study done by Sociopol. Consulted October 14, 2024. https://fcfa.ca/wp-content/uploads/2022/04/2022-02_Rapport-FCFA-Sociopol-cibles.pdf .

Government of Canada (2023, June 5). Action Plan for Official Languages – 2018-2023: Investing in Our Future. Consulted October 14, 2024. https://www.canada.ca/en/canadian-heritage/services/official-languages-bilingualism/official-languages-action-plan/2018-2023.html.

Houle, R. & Corbeil, J.-P. (2017). Language Projections for Canada, 2011 to 2036. Thematic Series on Ethnicity, Language and Immigration. Statistics Canada catalogue number 89-657-X. https://www150.statcan.gc.ca/n1/pub/89-657-x/89-657-x2017001-eng.htm.

Lutz, W. (2013). Demographic Metabolism: A Predictive Theory of Socioeconomic Change. Population and Development Review, 38(supplement), 283-301. https://doi.org/10.1111/j.1728-4457.2013.00564.x .

Lutz, W. (2021). Advanced Introduction to Demography. Northampton, MA, USA: Edward Elgar Publishing.

Morency, J.-D., Caron Malenfant, É., Coulombe, S. & Langlois, S. (2015). Projections of the Aboriginal Population and Households in Canada, 2011 to 2036. Statistics Canada catalogue number 91-552-X. https://www150.statcan.gc.ca/n1/pub/91-552-x/91-552-x2015001-eng.htm.

Morency, J.-D., Caron Malenfant, É. & MacIsaac, S. (2017). Immigration and Diversity: Population Projections for Canada and its Regions, 2011 to 2036. Statistics Canada catalogue number 91-551-X. https://www150.statcan.gc.ca/n1/pub/91-551-x/91-551-x2017001-eng.htm.

O’Donnell, V. & LaPointe, R. (2019). Response mobility and the growth of the Aboriginal identity population, 2006-2011 and 2011-2016. National Household Survey: Aboriginal Peoples. Statistics Canada catalogue number 99-011-X. https://www150.statcan.gc.ca/n1/pub/99-011-x/99-011-x2019002-eng.htm .

Ryder, N. (1965). The Cohort as a Concept in the Study of Social Change. American Sociological Review, 30(6), 843-861. https://doi.org/10.2307/2090964 .

Spielauer, M. (2010). Persistence and change in the relative difference in educational attainment by ethnocultural group and gender in Canada. Vienna Yearbook of Population Research, 8, 261-282. https://www.jstor.org/stable/23025517 .

Spielauer, M. (2014). The relation between education and the labour force participation of Aboriginal Peoples: A simulation analysis using the Demosim population projection model. Canadian Studies in Population, 41(1-2), 144-164. https://doi.org/10.25336/P6RG79 .

Statistics Canada (2016). Population and Family Estimation Methods at Statistics Canada. Catalogue number 91-528-X. https://www150.statcan.gc.ca/n1/pub/91-528-x/91-528-x2015001-eng.htm .

Statistics Canada (2021, October 6). Projections of the Indigenous populations and households in Canada, 2016 to 2041. The Daily. Consulted October 14, 2024. https://www150.statcan.gc.ca/n1/daily-quotidien/211006/dq211006a-eng.htm.

Statistics Canada (2022a, May 18). Microsimulation. Consulted October 14, 2024. https://www.statcan.gc.ca/en/microsimulation/index .

Statistics Canada (2022b, September 8). Canada in 2041: A larger, more diverse population with greater differences between regions. The Daily. Consulted October 14, 2024. https://www150.statcan.gc.ca/n1/daily-quotidien/220908/dq220908a-eng.htm.

Statistics Canada (2022c, August 22). Population Projections for Canada, Provinces and Territories, 2021 to 2068, 2022. The Daily. Consulted October 14, 2024. https://www150.statcan.gc.ca/n1/daily-quotidien/220822/dq220822b-eng.htm.

Statistics Canada (2022d, February 9). Canada tops G7 growth despite COVID. The Daily. Consulted October 14, 2024. https://www150.statcan.gc.ca/n1/daily-quotidien/220209/dq220209a-eng.htm.

Statistics Canada (2024a). Table 14-10-0327-01 Labour force characteristics by sex and detailed age group, annual. Consulted September 5, 2024. https://doi.org/10.25318/1410032701-eng .

Statistics Canada (2024b). Indigenous Peoples Technical Report, Census of Population. 2021 Census. Catalogue number 98-307-X. https://www12.statcan.gc.ca/census-recensement/2021/ref/98-307/index-eng.cfm.

United Nations (2000). Replacement Migrations: Is it a solution to declining and aging populations? Population Division.

United Nations Economic Commission for Europe (2018). Recommendations on Communicating Population Projections. United Nations.

van Imhoff, E., & Post, W. ( 1998).  Microsimulation Methods for Population Projection. Population: An English Selection, 10(1 ), 97- 138. http://www.jstor.org/stable/2998681.

Vézina, S., Morency, J.-D., Martel, L. & Pelletier, F. (2024). Canadian labour force: What will happen once baby boomers retire? Insights on Canadian Society, 1-12. Statistics Canada catalogue number 75-006-X. https://www150.statcan.gc.ca/n1/pub/75-006-x/2024001/article/00005-eng.htm.

Willekens, F. (2009). Continuous-time microsimulation in longitudinal analysis. In Ann Harding, A. Z. (Ed.), New Frontiers in Microsimulation Modelling, 353-376. London: Routledge.

Willekens, F. (2011). La microsimulation dans les projections de population. Cahiers québécois de démographie, 40(2), 267-297. https://doi.org/10.7202/1011542ar .

Wolfson, M. (1987). Homemaker Pensions and Lifetime Redistribution. International Association for Research in Income and Wealth. Rome, Italy. Consulted October 14, 2024. https://publications.gc.ca/collections/collection_2017/statcan/11f0019m/CS11-0019-3-eng.pdf.

Wolfson, M., Gribble, S., Bordt, M., Murphy, B., Rowe, G. & Scheuren, F. (1989, May 1). The social policy simulation database and model: an example of survey and administrative data integration. Annual meeting of the American Statistical Association, August 1988. New Orleans. Consulted October 14, 2024. https://www.thefreelibrary.com/The social policy simulation database and model: an example of survey...-a07724459 .

Wolfson, M. (1994). POHEM: a framework for understanding and modelling the health of human populations. World health statistics quarterly, 47(3/4), 157-176. https://doi.org/10.1186/s12963-015-0057-x .

Wolfson, M. (1995). Socio-Economic Statistics And Public Policy: A New Role For Microsimulation Modeling. Statistics Canada Working Paper, 81. https://www150.statcan.gc.ca/n1/pub/11f0019m/11f0019m1995081-eng.pdf.

Zaidi, A. & Rake, K. (2001). Dynamic microsimulation models: a review and some lessons for SAGE. SAGE Discussion Paper, 2. ESRC SAGE Research Group, The London School of Economics, London.