1.3. Introduction

Population projections are key for policy making and planning. Changes in the size and composition of the population are key determinants of the demand for goods and services, from basic food and education to energy and housing. Population projections, based on multiple scenarios, help government and other decision makers make informed decisions.

Most countries and international organizations, like the World Bank and the United Nations, produce population projections using the cohort-component method, a macro approach limited to a very small number of characteristics. This provides projections by age and sex at the national level, often disaggregated by urban/rural areas; large countries sometimes provide projections at a sub-national level. Given the high importance of education on human capital and its influence as the “single most important variable besides age and sex” on demographic behaviors (Lutz et.al. 1999), population projections that include education later became available for most countries, but that extension defined the technical limit of the cohort-component approach.

A more advanced—but still uncommon—approach consists of dynamic micro-simulation modeling, in which populations are represented by large samples of individual people whose life-courses is modeled over time. This approach is more complex, but has major advantages: it produces detailed projections of a broad variety of individual characteristics, models realistic life-courses and their diversity, and supports the modeling of interactions between people. The idea is to start with a micro-data population and make it evolve over time, unlike the cohort-component approach, which starts with a simple distribution of a population by age and sex but cannot track individuals over time. The idea is not new (van Imhoff and Post, 1998), but it has become feasible and affordable only recently with advances in freely available programming technologies and improvements in data availability.

Micro-simulation models allow more disaggregation and thus can provide projections of the population or some of its components, e.g., population by ethnic affiliation, school-age population by region, labor force by education level, etc. In addition, population projection micro-simulation models can provide the foundation for more specialized micro-simulation models of diseases, tax benefit, pensions, etc. For example, a micro-simulation population projection model is used by Statistics Canada to project the diversity of the Canadian population by visible minority group (Caron-Malenfant et.al. 2010), Aboriginal identity (Morency et.al. 2015), and labor force (Martel et.al. 2011), as well as to study the effect of educational improvements on the future size and composition of the Aboriginal labor force (Spielauer 2014). Dynamic microsimulation models for population projections in 27 European Countries also became available recently (Marois et.al. 2017).

Population projections by micro-simulation can be seen both as a complement to macro-projections and as their replacement, as macro-projection models can be implemented as micro-simulations, producing identical results, but allowing for model extensions beyond the technical limitations of cohort-component models. This point is demonstrated in this report with a micro-simulation example developed in two phases. The first phase reproduces the macro model DemProj (Stover & Kirmeyer 2001), which is widely used, especially in developing countries. In the second phase, the model is extended to incorporate more detailed characteristics and behaviors. Extensions include the fertility module, which allows a realistic projection of family sizes and fertility by education. We also added a module for child mortality, incorporating factors such as the mother’s education and age. Education is modeled to include its inter-generational transmission; first union formation is introduced as key determinant of the timing of first births.

This report demonstrates the feasibility and relevance of dynamic micro-simulation in the context of developing countries. The DYNAMIS-POP model is key to this report. While non-confidential synthetic data are provided for an imaginary country for demonstration and testing, the report also includes detailed instructions for its adaptation and replication in real countries. The model is implemented in Modgen, a freely available software package developed and maintained at Statistics Canada.

The report also introduces micro-simulation programming and various statistical methods used in micro-simulation, and includes step-by step documentation of its computer implementation. The model has an intuitive graphical user interface and runs on a standard PC. Its code and statistical analysis files are openly accessible and can be used for micro-simulation model development and implementation. New software solutions, which are free, and improved computer capabilities make it an affordable and much more feasible option for developing countries. Given the growing interest in disaggregated data for development planning and monitoring (for example to monitor the Sustainable Development Goals), it makes sense to also push for more disaggregation in projections and simulations. Besides producing more detailed projections, micro-simulation allows a more explicit incorporation of theory and policy levers in its projections. For example, it allows modeling of inter-generational dynamics and analysis of downstream effects on demographic change or child mortality of policy interventions that improve education.

This report has nine parts and is further accompanied by extensive materials including statistical analysis scripts, a step-by-step implementation guide, and model downloads. The seven parts of the report are:

  • An introduction to Dynamic Micro-Simulation for Population Projections (And Beyond)
  • The Philosophy and Goals of DYNAMIS-POP
  • Data Requirements to adapt the model to a new country context.
  • A description of the Modules of DYNAMIS-POP. Following the two-phase model development, we organize modules in those required to reproduces a typical macro model and those added in a second phase adding variables and processes beyond the macro framework to illustrate key features and strengths of the micro-simulation approach. The result is both a fully functional model that can be customized for other countries, and an illustration of typical modeling approaches found in micro-simulation. The model can be understood as a modeling platform, allowing for extensions.
  • A description of the Model Output including tables, micro-data files, and tracking databases.
  • Instructions on how to run the model using its Graphical User Interface
  • An introduction to typical Statistical Methods used in micro-simulation
  • A short discussion of the Modgen technology
  • Instructions on how the model can be ported to a new country.

This report is accompanied by three sets of documented code for analysis and model implementation.

  • A Set of R scripts for Parameter Estimation
  • A set of R scripts for statistical post-processing of micro-data and table output.
  • A Step-by-Step Implementation Guide including all programming code. This part discusses technological aspects of building micro-simulation models in detail. It covers the technical implementation of the model from a developer’s perspective, thereby introducing the key components of the Modgen programming technology and resulting applications.

DYNAMIS-POP is fully reproducible. All statistical analysis scripts (R code) used to generate the model parameters as well as the model code are openly available for download.

Data requirements are met by most countries through their population censuses, and implementation of demographic household surveys like UNICEF’s Multiple Indicators Cluster Surveys (MICS) or Demographic and Health Surveys (DHS). Once built and compiled, using the model is easy, as it has an intuitive graphical user interface and runs on a standard PC. While model development remains a complex issue, a collection of modular components that can be adapted and assembled for various purposes is being built. While it will not be a plug-and-play tool and still require technical expertise for model development or customization, this collection of well-documented modules should save developers and data analysts considerable time and lower the entry barrier into micro-simulation modeling.

As the model remains in development, upgrades of existing modules, and additional modules, are expected to be released in the future. All materials are made public on the project Github repository.