Proceedings of the 2014 Winter Simulation Conference
A. Tolk, S. Y. Diallo, I. O. Ryzhov, L. Yilmaz, S. Buckley, and J. A. Miller, eds.
The Role of Languages for Modeling and Simulating Continuous-Time Multi-Level Models in
Demography
Alexander Steiniger
Adelinde M. Uhrmacher
University of Rostock
Albert-Einstein-Str. 22
D-18059 Rostock, GERMANY
Sabine Zinn
Jutta Gampe
Frans Willekens
Max Planck Institute for Demographic Research
Konrad-Zuse-Str. 1
D-18057 Rostock, GERMANY
ABSTRACT
Demographic microsimulation often focuses on effects of stable macro constraints on isolated individual
life course decisions rather than on effects of inter-individual interaction or macro-micro links. To change
this, modeling and simulation have to face various challenges. A modeling language is required allowing
a compact, succinct, and declarative description of demographic multi-level models. To clarify how such
a modeling language could look like and to reveal essential features, an existing demographic multi-level
model, i.e., the linked life model, will be realized in three different modeling approaches, i.e., ML-DEVS,
ML-Rules, and attributed pi. The pros and cons of these approaches will be discussed and further requirements
for the envisioned language identified. Not only for modeling but also for experimenting languages can
play an important role in facilitating the specification, generation, and reproduction of experiments, which
will be illuminated by defining experiments in the experiment specification language SESSL.
1 INTRODUCTION
In demography, in addition to macro-projection (Bohk et al. 2009), microsimulation plays an important role
(Willekens 2005, Willekens 2009, van der Gaag 2009). Its essence is the individual’s life-course, which is
defined by the sequence of states that the individual visits over time and the waiting times between state
transitions. The time-scales can either be discrete (usually in units of years) or continuous. In discrete-time
models, time advances in discrete, equidistant steps. In contrast, a continuous-time microsimulation has a
continuous time scale along which events can occur. In general, continuous-time models are the optimal
theoretical choice for a precise description of population dynamics, as they mirror life-course developments
most closely (Willekens 2005). The underlying stochastic model of continuous-time microsimulation is
a continuous-time Markov model parameterized with transition rates. Accordingly, a continuous-time
microsimulation model is also a competing risks model (cf. Zinn (2011)).
Currently, demographic microsimulation focuses often on effects of stable macro constraints on isolated
individual life course decisions rather than on effects of inter-individual interaction (Jager and Janssen 2003)
or the macro-micro link (Ewert et al. 2007, Gilbert and Troitzsch 2008). To change this, we have to face
various challenges. Two issues seem to refer to modeling: (a) how to model inter-individual interactions
and dynamic binding/grouping (e.g., household formation and dissolution) in continuous space and time
and (b) how to integrate dynamics at different levels. Another challenge refers to validation, as with
more expressive models, demography has started to move from data- to hypotheses-driven development of
individual-based, mechanistic models. This has an impact on validation processes and methods used. The
question how far already existing approaches address requirements of multi-level modeling and simulation
in demography, we will explore based on the linked lives model of Noble et al. (2012).
2978 978-1-4799-7486-3/14/$31.00 ©2014 IEEE