AN APPROACH TO COMPARATIVE POPULATION MODELING AND SIMULATION
Sunday, January 10, 2016: 11:15
Kai Chong Tong Auditorium, G/F (Jockey Club School of Public Health and Primary Care Building at Prince of Wales Hospital)
Purpose: Simulating aspects of the health system, such as number of diseases, people needing a treatment, or nation-wide costs often require a valid representation of the population. We present two population models that can be used as a basis for simulations in the health system and show how they can be parameterized based on given data and simulate the population accurately.
Method(s): We developed two different models for the simulation of a population: an agent-based model, which simulates individuals over discrete time steps, and a system dynamics model, which simulates aggregates that represent population groups over a continuous time. Both models include the characteristics age and gender, as well as births, deaths, immigrations and emigrations, and both models and are designed to simulate a changing population over time. In a first step, we defined model structures for both models to incorporate the population characteristics valid and accurately. In a second step, we gathered data from Statistics Austria, including assumptions about birth, death, immigration, and emigration rates until 2076. This data was used to calculate parameter values. The computations rely on statistical methods, mostly for aggregation and computing probabilities over specific intervals or for continuous changes. In a third step, the Austrian population is simulated until 2076.
Result(s): Results: The two models simulate the same system but their structures and parameters are fundamentally different. The computations of all model parameters are possible. Results are presented as the population and its demographics for each year. As the prognostic model starts in 2000, it was possible to validate the first 15 years of the simulation with real aggregated data, also gained from statistics Austria. Both model results only differ by less than one percent from the real population in 2015 as well as from the prognosis for 2076.
Conclusion(s): In our comparative analysis, both modeling methods are eligible to model and simulate populations over time. Small deviations are caused by structural model differences. For example, there are different ways to define and compute the mean population in a year. In our explorative example, the differences are small enough to accept both results as correct. For further simulation studies, this allows to integrate the population in a standardized, valid way if one of the two methods is used.
Method(s): We developed two different models for the simulation of a population: an agent-based model, which simulates individuals over discrete time steps, and a system dynamics model, which simulates aggregates that represent population groups over a continuous time. Both models include the characteristics age and gender, as well as births, deaths, immigrations and emigrations, and both models and are designed to simulate a changing population over time. In a first step, we defined model structures for both models to incorporate the population characteristics valid and accurately. In a second step, we gathered data from Statistics Austria, including assumptions about birth, death, immigration, and emigration rates until 2076. This data was used to calculate parameter values. The computations rely on statistical methods, mostly for aggregation and computing probabilities over specific intervals or for continuous changes. In a third step, the Austrian population is simulated until 2076.
Result(s): Results: The two models simulate the same system but their structures and parameters are fundamentally different. The computations of all model parameters are possible. Results are presented as the population and its demographics for each year. As the prognostic model starts in 2000, it was possible to validate the first 15 years of the simulation with real aggregated data, also gained from statistics Austria. Both model results only differ by less than one percent from the real population in 2015 as well as from the prognosis for 2076.
Conclusion(s): In our comparative analysis, both modeling methods are eligible to model and simulate populations over time. Small deviations are caused by structural model differences. For example, there are different ways to define and compute the mean population in a year. In our explorative example, the differences are small enough to accept both results as correct. For further simulation studies, this allows to integrate the population in a standardized, valid way if one of the two methods is used.