Title: Simulation Methods for Longitudinal Analysis
Venue and date: Universiteit Hasselt (formerly Limburgs Universitair Centrum), Hasselt, Belgium , 6-14 October 2005
Hosted and co-sponsored by the Centre for Statistics, Universiteit Hasselt
Senior Instructors:
Dr. Geoff Rowe, Socio-Economic Analysis and Modeling Division, Statistics Canada , Ottawa, Canada
Dr. Frans Willekens, Netherlands Interdisciplinary Demographic Institute (NIDI), The Hague and Population Research Centre, University of Groningen, Netherlands
Course Assistants:
Dr. Chantal Hicks, Senior Research Analyst, Socio-Economic Analysis and Modeling Division, Statistics Canada, Ottawa, Canada
Dr. Corina Huisman, Researcher, Netherlands Interdisciplinary Demographic Institute (NIDI), University of Groningen, Netherlands
Workshop Summary:
Life is an evolving process (staging process) embedded in a complex context. The aim of dynamic simulation is to capture that process in a dynamic model that can be used to assess the impact on the lives of people of events, policies and other interventions. Life histories are characterized in the following general way: (1) at any point in time or age, an invidual is characterised by a finite number of attributes or states; (2) changes in attributes (or transitions or events) may occur at any point in time; and (3) there are time-constant and/or time-varying factors influencing the timing of events. Examples of attributes are marital status, living arrangement, health status, employment status, and region of residence. A change in attribute involves a movement from one state (state of origin) to another (states of destination). The rate of movement or the rate of transition may depend on personal attributes, including conditions and experiences earlier in life, and on the context, including institutional factors.
This workshop focused on models to generate 'synthetic biographies', that are life histories obtained by combining partial information on the life histories of different people. Rates and probabilities of transition between states are estimated from fragmentary data on individual life histories and they are applied to produce entire biographies that are consistent with the data. The models are known as multistate models. The models describe how the attributes of people change as they age, when the changes occur and how long people spend in the various stages of life. Policies and other interventions affect the lives of people. Some effects are immediate while other are delayed. The models are instruments for dynamic impact assessment. Life histories can be generated at the population level, at the cohort level (cohort biographies) or at the individual level (individual biographies). Sometimes a distinction is made between macro-simulation (population and cohort level) and micro-simulation (individual level).
The workshop covered the design and estimation of multistate models and their use in the simulation of cohort biographies, individual biographies. Areas of application include health sciences, sociology, demography and economics.
The workshop consists of two parts:
• LifePaths developed by Statistics Canada. Canada is the world leader in developing new tools for policy planners. LifePaths creates a virtual nation of synthetic Canadians, whose characteristics mirror those of real Canadians. The model reconstructs the life courses of those synthetic individuals by combining the many fragments of real lives that are captured by existing surveys and administrative records (Rowe (2004) Fragments of lives: enabling new policy directions through integrated life-course data http://policyresearch.gc.ca/page.asp?pagenm=v6n2_art_03)
• Biographical models: multistate models for generating synthetic biographies. This part discusses the mathematical and statistical theory of multistate models, including multistate life tables and projection models. Multistate life tables describe the experience of synthetic cohorts. Projection models relate to real cohort. The projection package LIPRO will also be presented (http://www.nidi.knaw.nl/en/projects/section2/270101/ )
Preparatory Reading
In addition to the reading list distributed via email, participants read the following article before coming to the workshop:
The LifePaths Microsimulation Model: An overview
Supplementary Reading
Some of them were referred to in the course of the workshop:
Privatizing Public Pensions: Canadian Analyses
Rate of Return to Education - Distributional Analysis using LifePaths
Chapter 8 of Government Finances and Generational Equity
Longitudinal Analysis of Labour Force Survey Data 1
Longitudinal Analysis of Labour Force Survey Data 2 (a more recent version of this paper is available upon request. Please email the Programme Coordinator.)
Seminar Presentations:
- Strategies to Fit Pattern-Mixture Models, Herbert Thijs et al., Biostatistics, Center For Statistics, Limburgs Universitair Centrum, Universitaire Campus; Janssen Research Foundation; Biostatistical Centre, Katholieke Universiteit Leuven; and EORTC Data Centre, Belgium
- Consultant's Forum - A Local Influence Approach Applied to Binary Data from a Psychiatric Study, Ivy Jansen et al., Biostatistics, Center For Statistics, Limburgs Universitair Centrum, Universitaire Campus, Belgium
- Kernel weighted influence measures, Niel Hens et al., Center For Statistics, Limburgs Universitair Centrum, Universitaire Campus; and Biostatistical Centre, Katholieke Universiteit Leuven, Belgium
- The nature of sensitivity in monotone missing not at random models, Ivy Jansen et al., Center For Statistics, Limburgs Universitair Centrum, Universitaire Campus; Biostatistical Centre, Katholieke Universiteit Leuven, Belgium; and Medical Statistics Unit, London School of Hygiene and Tropical Medicine, UK